The Republic of Panamá is located at approximately 8–9° N latitude (Fig. 1), which is a latitude central to the native range of Saccharum spontaneum (Mukherjee 1957). Day length varies only about one hour from its maximum at the June summer solstice to December and average daily temperatures range between 19°C and 35°C. During the rainy season months of May to December, rainfall averages around 240 mm per month, with rainfall occurring frequently (World Meteorological Organization 2010). All studies occurred during 2009, which was a year with typical rainfall patterns.

We collected seeds of Saccharum spontaneum from 12 sites (Sites 1–12 Fig. 1; Appendix 1) in the PCW weekly between September and December. Each week three mature inflorescences, from different plants within a few meters of each other, were sampled and assayed separately. Seeds that had disarticulated from the inflorescence but remained loosely attached were collected, and kept at room temperature until assayed within two days of harvest. When wet at the time of harvest due to rain, seeds were dried under ambient conditions prior to conducting germination assays.

One hundred seeds were taken at random from those collected from each inflorescence and germinated on moist paper towel in petri dishes (n=3 inflorescences per site per week). Seeds were germinated at 36°C as Saccharum seeds have a very high temperature optimum for germination of around 36°C (Skinner 1959, Singh 1988). Assays were conducted in the dark, as there is no evidence of a light requirement for the germination of Saccharum seeds (G. Bonnett, unpublished).

Germination was defined as root or radicle emergence visible to the naked eye. Seed germination was checked every three to four days and counted seedlings were removed from the plates. Total number of seeds germinated after two weeks is presented. For analysing the effect of date of collection on germination percentage we carried out a one-way ANOVA with date as a factor and log-transformed (log10(value+1)) germination data as the dependent variable (Sigmaplot 11, Systat Software Inc. San Jose, CA). Post –hoc comparisons of means were conducted using the Tukey method, for this and all subsequent analyses significance was tested at the 0.05 level.

To test if seeds would germinate in situ, vegetation was cleared and the soil surface raked on 16 September exposing areas of bare soil between two Saccharum spontaneum stands (Site 6, Fig. 1, Appendix 1). The experiment comprised 10 replicates of three treatments, each in 60 by 120 cm plots. Treatment one, bare soil, was designed to observe any seedlings emerging from seeds deposited by wind. The second treatment comprised six mature inflorescences placed on each plot and covered with nylon flywire to prevent seeds from blowing away. Plots in treatment three were bare soil covered by nylon flywire, with no additional seeds added. One replicate of each treatment was randomly placed in each of 10 evenly spaced blocks. After 19 days we removed the mesh to allow continued growth of seedlings. The number of germinated Saccharum spontaneum seedlings per plot was recorded one week after, and then every two weeks until nine weeks after the start of the experiment.

To test for differences in seedling numbers between treatments, we conducted a repeated measures analysis with time (within subject) and treatment (between subjects) as factors and the number of seedlings as the dependent variable (Sigmaplot 11). The number of seedlings was log-transformed (log10(value+1)) prior to analysis and post –hoc comparisons of means were conducted using the Holm-Sidak method (a variation of the Holm multi-comparison test using the Sidak correction (Sidak 1967)).

To assess levels of genetic diversity within stands, plants were collected at 22 sites across central Panama, with a focus on the PCW but including the eastern- and westernmost edges of the invasion (Fig. 1, Appendix 1). Green leaves were harvested from individual culms spaced up to 10 m apart (n = 3 per site). To examine if the patterns we found at these sites were consistent across large stands, we also sampled parallel transects separated by 20 m at two independent stands in Soberania National Park within an area dominated by Saccharum spontaneum (Sites 5 and 6, Fig. 1, Appendix 1; see Saltonstall and Bonnett 2012 for a description of the stands). Five transects were laid out in Stand 1 and three in Stand 2 and samples collected every 10 m along the transects (n=26 for Stand 1, n=9 for Stand 2). All samples were kept frozen at -20°C until DNA was extracted using a CTAB extraction protocol (Doyle and Doyle 1987).

We assessed multilocus allele profiles using 12 microsatellite loci (msscir14, msscir17, msscir53, msscir58, SMC28, SMC221, SMC334, SMC336, SMC597, SMC1047, SMC1237, SMC1493; Brown et al. 2007; Pan 2006) targeting different regions of the genome. PCR amplifications were performed in 7 μL reaction volumes containing approximately 1 ng DNA, 0.056 U Amplitaq (Applied Biosystems, Inc., Carlsbad, CA), 1x reaction buffer (Applied Biosystems, Inc.), 3 mM MgCl₂, 200 uM dNTPS, 0.5 pmol of forward primer including an M13 tag, 2.5 pmol of reverse primer, 2 pmol of FAM, PET, VIC, or NED labelled M13 primer, and 1 mg mL^-1 BSA. All PCR reactions were conducted with 2 min of denaturation at 94°C, 35 cycles of denaturation at 94°C for 45 s, annealing at 56°C for 25 s, and 25 s of extension at 72°C, and a final extension period of 72°C for 5 min in a thermal cycler (Eppendorf Multimax, Eppendorf, Hamburg, Germany). Primers were not multiplexed due to the high ploidy level of the species, but PCR reactions with differing florescent tags were multiloaded for capillary electrophoresis on an ABI prism 3730 automated sequencer. Fragment sizes were analysed using GeneMapper 3.7 (Applied Biosystems Inc.) and individuals were scored for presence or absence of identified alleles for each locus. All scored alleles had consistent results over replicate amplifications. Due to polyploidy in this species, allele frequencies are unknown thus multilocus allele profiles are hereafter referred to as allele phenotypes. We compared all multilocus allele phenotypes found within and among sites to identify repeated allele phenotypes, which we assumed to result from asexual reproduction (e.g. clonal expansion).

Effect of drying: To assess the ability of buds to remain viable and sprout after different periods of drying, culms were collected from Site 4 (Fig. 1, Appendix 1) on 3 September. Culms, approximately 2 m tall, that had not flowered were cut at the base and randomly assigned to five treatments (drying durations) spread out on benches that were shielded from rainfall but otherwise open. Drying durations were 0 days (day of harvest), one, two, four, and six weeks. After drying, stem pieces with a node at either end were cut from the culms and ten were placed into each of five replicates and covered with commercial peat. Each replicate comprised two plastic seedling trays (53 cm by 26 cm). Plants were grown under an open structure with a transparent roof, watered daily, and the plants that sprouted recorded four weeks after planting. To assess changes in moisture content, plant material was harvested from the same site on 19 November. Fresh and dry weights of stem pieces cut from culms at the time of harvest and after one, two and four weeks were recorded. Dry mass was taken after drying the stem pieces to constant weight at 40°C.

Data from the different times of drying were compared by ANOVA and times of drying were compared with Tukey multiple comparison tests (Sigmaplot 11). The proportion of buds that sprouted was square root transformed, other data was not transformed prior to analysis.

Fresh culms and rhizomes were collected at Site 6 (Fig. 1, Appendix 1) on July 3 and propagules were cut to the appropriate size and planted the same day. Both experiments were performed under a covered shelter, open on the sides. Unmodified soil representative of the area (Oxisol, collected in Rio Hato, Panama) was used and all pots were watered daily or as needed. All propagules were harvested on September 12 when the presence of a sprout above-ground, a sprout below-ground but not emerged, or the presence of roots was recorded.

Propagule size: Culm pieces with one node (and bud) were cut into either 2, 5, 10, or 20 cm lengths or two node (and bud) pieces ranging in length from 9.9 to 12.7 cm. Rhizomes were cut into pieces ranging from 15–40 mm in length, either with or without visible axillary buds. There were 10 replicates for each culm length (single and two node pieces) and 25 pieces of each rhizome type. Culm pieces shorter than 10 cm were planted individually in 10 cm diameter plastic nursery sacks while longer pieces were planted in 12 L pots. Rhizomes were planted individually in 10 cm sacks. All propagules were planted at a depth of 5 cm.

Planting depth: Individual propagules (5 cm single node stem pieces or single node rhizome pieces (ranging from 11–44 mm in length)) were planted at different depths in 10 cm nursery sacks. Planting depths were surface, 5 cm, 10 cm, and 20 cm for stem pieces, and surface, 10 cm, and 20 cm for rhizome pieces; there were ten replicates per planting depth. Time to sprouting was recorded for each propagule.

Our original intent was to analyse growth of sprouts using analysis of variance appropriate to the experimental design. However, because of the frequency of failure to sprout, and thus zero growth, we decided to analyse only the proportion sprouted either above- or belowground, as the percentage of the 10 replicates that sprouted for each treatment.

To test if soil type influenced the results of this experiment, growth was also tested using commercial potting mix in nursery sacks. Single node stem pieces (5 cm) were planted at the surface or 2 cm depth, in either commercial potting mix or unmodified soil from Rio Hato, with six replicates per treatment. Presence of sprouting was recorded after four weeks.

We further distinguish between propagules that produced roots at the node but did not initiate other growth, propagules that sprouted but died before they reached the soil surface, and propagules that produced visible growth above the soil surface. Results were analysed using generalized linear models (Poisson GLM for propagule depth and Binomial GLM for propagule size and soil type) as implemented in R 2.13.1 (R Development Core Team 2010). The proportion of variance explained by each model (r²) was calculated from the null and residual variances.

Germination ability

The proportion of seeds germinating and the temporal patterns of germination varied greatly across the 12 sites (Appendix 2). The range of maximum germination percentages were 72%–30% (Appendix 2), with an average level of germination around 35% across sites until November (Fig. 2). Significantly fewer seeds germinated after 16 November (ANOVA, F_{12, 403} = 25.5, P<0.001, Fig. 2).

In situ germination

The numbers of seedlings germinating in situ showed a significant effect of treatment (F_{2, 18} = 21.6, p<0.001), week (F_{4, 36} = 7.4, p<0.001) and treatment by week interaction (F_{8, 72} = 29.9, p<0.001), so the effect of treatment was dependent upon time. One week after the experiment was initiated, germination was evident in all plots (Fig. 3) and there were significant differences in the number of seedlings between each treatment, with the plots with added seeds having many more seedlings than the controls. The average numbers of seedlings in plots with added seeds decreased from week to week, while those in control plots increased over time. Within the flywire treatment, there was a significant increase in the number of seedlings between weeks 3 and 5, corresponding with the removal of the flywire, whereas the seeds added and bare soil control plots did not show significant changes after week 3. After nine weeks there were no significant differences in the numbers of seedlings remaining between any of the treatments.

Genetic diversity of mature stands

High levels of genetic diversity were found both within and across sites. A total of 227 alleles were observed across the twelve microsatellite loci, of which 215 were shared between sites. The average number of alleles per locus (A_o) was 18.4 and up to ten alleles per individual were found within a locus (range 1-10).

In the two stands sampled with transects, the majority of culms sampled were unique (Stand 1 at Site 5 = 89% and Stand 2 at Site 6 =65%) and all replicate allele phenotypes were found in adjacent plots. Similarly, in the 22 sites where only 3 samples were collected, 86% of them contained multiple allele phenotypes (19 of 22 sites). Fifty four multilocus allele profiles were found across the 22 sites, with three unique phenotypes sampled at each of 13 sites (Sites 3, 5, 6, 11, 13-16, and 18-22), two phenotypes at each of six sites (2, 4, 7, 9, 10, and 17), and a single phenotype at each of three sites (1, 8, and 12; Appendix 1). The same multilocus profile was not found at more than one site.

Effects of drying

There was a significant effect of time of drying of the culm on the proportion of buds that sprouted (ANOVA F_{4, 19} = 16.992, P<0.001, Table 1), less than half of the buds sprouting after 4 weeks. However while significant over time, the reduction in moisture content was only 8% lower after 4 weeks of drying (ANOVA F_{3, 16} = 7.819, P<0.01, Table 1).

Effects of propagule size

Increased size of stem fragments increased sprouting ability (p<0.01, r² = 0.34; Fig. 5). While roots were produced in the majority of fragments of 5, 10 and 20 cm lengths, no fragments of 2 cm, 5 cm, or 10 cm produced active sprouts from the node. One 20 cm (10%) and one 2-node fragment (10%) sprouted belowground but died before reaching the soil surface. Three (30%) 20 cm fragments and seven (70%) 2-node fragments sprouted aboveground. All 2-node fragments that grew sprouted at only one node, but had roots present at both nodes. No rhizome fragments, with or without buds, sprouted.

Effects of planting depth

There was a significant negative effect of planting depth on the ability of stem cuttings to sprout, with lower depths having an inhibitory effect on sprouting (p<0.05, r² = 0.29, Fig. 4). While 100% of stem cuttings laid at the soil surface sprouted, only 40% successfully grew aboveground from a depth of 5 cm and only 20% from 10 cm. Stems at a depth of 20 cm failed to grow aboveground, but 20% of propagules at this depth sprouted and died before reaching the soil surface. All rhizome pieces, at any depth, failed to sprout suggesting that single node pieces of Saccharum spontaneum rhizomes may have low sprouting ability (data not shown). One rhizome piece, at 10 cm depth, produced roots but did not produce a sprout. Rhizome pieces laid at the soil surface dried out quickly, despite daily watering, and did not even produce roots.

Depth trials comparing commercial potting soil (PS) to the unmodified soil (US) used in the previous trial showed similar results, with 83.3% (PS) and 67.7% (US) of cuttings laid at the soil surface sprouting aboveground and 67.5% (PS) and 50.0% (US) sprouting when planted at a depth of 2 cm, respectively. No significant differences were seen between depths or soil types (p>0.05, r² = 0.05).

Management of Saccharum spontaneum in Panama is restricted to controlling aboveground biomass, and generally involves physical cutting of mature biomass, and in some cases, chemical control where reforestation is a goal (Craven et al. 2009). Fire is also used during the dry season months to remove biomass. When cut, biomass is typically left on site to dry and degrade naturally. These activities occur year round, and regrowth of stands following biomass removal occurs frequently and rapidly (Saltonstall pers. Obs). Like many other weeds (Holm et al. 1977), the remaining buds and rhizome system thus presents the greatest barrier to control of established stands of Saccharum spontaneum. However, little thought goes into the effect of these management efforts on seed production and further spread of the species.

Timing and method of control thus become important when considering management of this aggressive invader in Panama. While intermittent flowering of Saccharum spontaneum occurs year round, the peak season of flowering is from August – October. Our studies have shown that seed germinability is highest in the months of September, October, and early November with a rapid decline in germinability in December. We suspect that germinability of seeds produced at other times of the year is also much lower. Timing of biomass removal can be important, with removal in June or July, prior to inflorescence emergence and seed development but late enough in the year that reproduction will not occur in the months of August – November, possibly being a method to reduce spread by seed. This may be particularly important in areas on the edges of the distribution where Saccharum spontaneum does not yet dominate and spread can be minimized on a local scale. As herbicides are not commonly used in Panama except during the initial stages of reforestation projects, repeated cutting of stems throughout the year to deplete below-ground carbohydrate resources in rhizomes is another approach which may help to prevent flowering at optimal times of the year. However, as our studies have shown that cut stems are still capable of sprouting after six weeks of desiccation, cut biomass should also be handled to minimize sprouting and spread to new sites.

Better control of dry-season fires is also needed to help control spread of Saccharum spontaneum by seed, as rapid regrowth following the onset of the rainy season typically leads to stems with mature infloresences developing in the peak months of seed production. Further, burned areas have higher densities of flowering stems and comparable seed germination rates as unburned areas (Saltonstall and Bonnett 2012) suggesting that the potential for spread to new sites is increased by fire.

As Saccharum spontaneum in Central Panama is widespread and many thousands of seeds are produced by each plant, a strategy to control seed numbers will have to be co-ordinated over a large area to be truly effective. While the extent of seed dispersal has not been experimentally quantified, research conducted on Barro Colorado Island (BCI) in the Panama Canal, utilising 200 seed rain traps (Wright and Calderón 2006), provides some evidence. Saccharum spontaneum is found only in small, isolated patches on BCI, mostly within treefall gaps in the forest interior (S.J. Wright, personal communication). However, between 1990 and 2011 over 7500 Saccharum spontaneum seeds were trapped (annual median 229); S.J. Wright unpublished data). These seeds were likely wind dispersed from mainland areas and would have travelled at least 2 km across the Canal from the nearest sources of Saccharum spontaneum plants. Information regarding the longevity of Saccharum spontaneum seed remaining in the seed bank is also lacking, but a concerted effort over several years to eliminate Saccharum spontaneum seed from the seed bank will be required to stop spread by seed. The closely related invasive Andropogon gayanus has been shown to retain 1% of germination of seed after one year in burial experiments (Flores et al. 2005), suggesting that establishment of new plants could continue to occur for some time even if annual seed production is reduced across the landscape.

In conclusion we have presented evidence that the spread of Saccharum spontaneum in Panama has been driven by seed production, but that vegetative propagules derived from stem fragments are also robust and can be the source of new plants. Seed production from Saccharum spontaneum can be reduced through the timing of management actions to reduce biomass that prevents re-growth of flowering stems. However the mobility of the seed means that to be effective, such actions would have to be conducted over large areas. Once established, the persistence of individual stands through re-growth from buds lower on the stems and rhizomes will continue to impede control and eradication over large areas so further research to determine a weak point in the vegetative growth stage is needed.