In fact, although mycorrhiza obviously could be an important cause of species distribution, their effect has never been studied, except in a fairly crude, qualitative way, relating types of mycorrhizal fungi to broad types of vegetation (tropical rainforest, temperate forest, heathland, grassland etc.). Therefore, this study (which attempts to quantify the effects of mycorrhizal fungi on a much more local scale) is unique.

Since this is a first study in this field, it is a fairly simple, general one. The mycorrhiza-vegetation system is treated as a 'black box'; we imagine the system as a piece of machinery in a closed box. We can measure what goes into the system (different intensities of mycorrhizal infection) and what comes out (different patterns of plant species), but we do not investigate the mechanisms within the box, merely relate the outputs to the input. This may subsequently allow us to make inferences about the workings of the box, but we cannot test these without further work (in effect, opening the box). Therefore the study is purely correlative; it seeks only to describe the way in which plant communities are affected by mycorrhiza, not how this happens.

For this purpose, the Svalbard tundra is ideal. Firstly, Svalbard has very few native plant species. Therefore, the system we are investigating is very simple, as are the patterns of species we may observe. This makes it much simpler to relate these patterns to different intensities of mycorrhizal infection, and minimises patterns caused by the interactions between plant species, or chance effects. Secondly, the possibilities that the pattern of the vegetation is due to either the action of herbivores or man may be rejected; Svalbard has no native rodents or rabbits, and very few insect grazers. In fact, the only grazing animals on the island are reindeer, and these were prevented from reaching the experimental site by a glacier. Similarly, the area is too remote to have been significantly disturbed by human activities (Pyramiden, the nearest settlement, was 25km away, and there are no roads). Finally, a variety of habitats, from sand dunes to scree, occur within a very small area, making it a convenient place to work.

In order to show that mycorrhiza are an important cause of plant distribution, it is necessary to exclude all other possible causes of patterning. The factors commonly thought to determine plant distribution include climate, soil factors and topography (Kershaw). Accordingly, these were measured at a number of sites in the study area, along with the abundance of mycorrhizal fungi in the soil and the nature of the vegetation at the site.

This data was then analysed by a statistical technique known as ordination analysis, using an Apple Macintosh computer. Ordination works as follows. Using the data collected on the species present at each site, a coefficient of dissimilarity may be calculated for each pair of sites. This coefficient has a value between 0 (identical vegetation at both sites) and 1 (complete dissimilarity; all different species). Now imagine that each site represents a point in space. This is not real space, but an imaginary mathematical space ('phase space') designed to show the relationship between the sites. The coefficients represent the distances between the sites. Therefore, the more similar the vegetation at two sites, the lower their coefficient of dissimilarity and the closer the two sites lie in phase space. The points will form an ellipsoidal cluster in space. The axes of the ellipsoid will represent the axes of variation between the sites; thus, the longest axis of the ellipsoid will represent the major axis of variation between the sites, and therefore the most important way in which the vegetation varies in reality. By mathematically determining these axes, each site may be assigned co-ordinates in phase space. By comparing the strength of the correlation between the co-ordinates of a site and the measurements of the environment at the site, it is possible to say which of these measurements best corresponds to the axes of variation. Thus, the strongest correlation between an environmental factor and the co-ordinates along the longest axis of variation indicates that this factor is the most important cause of the variation in the data, and therefore the most important cause of the patterns in the vegetation in the field.

METHODS AND MATERIALS

Site
The site lies at the foot of Alandsdalen, a glaciated valley running into the head of Austfjorden; the large fjord which divides the north of Spitsbergen into two (78^o 50' N 16^o10' E). The site is located on the lower slopes of the mountain Sentinelnosa, between steep scree above and dunes above the ice-dammed lake of Alandsvatnet below. It is sheltered on three sides by the mountains Sentinelnosa, Odellfjellet and Trikolorfjellet; in consequence the site is shaded, but rarely exposed to strong winds. During the fieldwork period, weather conditions were mild and stable, with daily mean temperatures ranging from 4^oC to 11^oC and extremes from 3^oC to 13^oC. Although precipitation was extremely low (2mm during the entire month), humidity was high (daily mean relative humidity: 55-93%) and fog was common (14 days out of 29). During this time, the climate became gradually more diurnal, although 24 hour daylight was present throughout.

The soil is derived from Devonian sedimentary rocks, and is nearly neutral in pH, although relatively poor in nutrients. The vegetation at the site is sparse (5-95% cover), and dominated by grasses and creeping plants. Mosses and terricolous (Soil living) lichens are also abundant and ubiquitous. Fieldwork was carried out between July 6^th and August 4^th 1993.

The site was chosen to represent the largest range of habitat types and associated vegetation possible. Thus, although the study area may seem limited in size, it includes a large altitudinal range of habitats, from the dunes above the lake, through more consolidated alluvial plains to the scree slopes above. A range of soil conditions (most notably soil water content, depth and texture) are also represented. The only major vegetation type not represented is the sparse flora of the high mountains. Therefore, the study may be taken as representative of a wide range of locations within Svalbard, although the milder climate of the western coast produces a slightly different flora.

Similarly, although the study was limited in time, the period chosen represents almost the whole of the summer growing season. During the other months of the year, the area is under snow and ground ice and the vegetation is dormant. Therefore, the causal factors in plant distribution at the site will be the conditions the vegetation experiences during this growing season alone.

Preliminary survey
During the first week of fieldwork, a square study area of 0.5km x 0.5km was marked out, and mapped at a scale of 1:5000 by trigonometric survey, using a grid of 100 evenly spaced survey points. Distances were measured by pacing and inclinations and bearings were measured with a compass clinometer. This area was further subdivided into a 10 by 10 grid, the sides of each smaller square being 50m. One sampling site was chosen within each smaller square by means of random co-ordinates, and marked with a cairn. Therefore, these sites provided a stratified random sample of conditions in the entire study area.

In addition, a full species list was compiled for the area, species being identified by reference to 'Flora of Svalbard' (Gjaerevoll & R¿nning, 1989) and 'Grasses, sedges, rushes and ferns of Britain and Northern Europe' (Fitter et al., 1990). Specimens of unidentified plants and lichens were preserved by drying, for comparison to the reference collection in Cambridge University Plant Sciences department herbarium.

DETAILED SURVEY

Vegetation
Percentage cover of every species of plant and lichen was measured at each survey site by means of point counting. A regular 10 by 10 grid of points was created by tying strings across a square 9m² quadrat at constant intervals, and the plant species lying beneath each intersection of the strings was recorded. The percentage of the area of the quadrat covered by a species is the same as the percentage of intersections under which the species lies.

Topography
Three measures of local topography were recorded at each site. Aspect and inclination were measured with a Silva compass clinometer, while altitude was determined by interpolation from the preliminary survey data. These were intended as indirect axes of patterning in the flora, representing of combination of climatic, edaphic and exposure factors.

In addition, exposure was quantified directly on a microtopographical scale. The method was a modification of the 'Topex' method; the immediate (to a distance of 10m) inclination of the site was recorded in all four compass directions, and the mean calculated. Negative inclination values were assigned to slopes descending below the site.

Climate
Climate measurements (and those of soil water content) were repeated three times over a period of three weeks in order to obtain both mean and range values. Further replicates would have been desirable, but were unfeasible in the time allowed. Measurements of each type were taken on the same day at every survey site (between 10.00 and 20.00) in order to maintain internal consistency in the data. Inter-site differences in diurnal variation were not measured, but were assumed to be negligible. Diurnal variation in climatic conditions was itself small relative to the day to day variation (±1^oC compared to ±8^oC).

Temperature and relative humidity were measured by means of a whirling sling hygrometer, while light climate was recorded as the number of hours of direct sunlight received at the site in one day. Measures of total radiation received would have necessitated continuous recording at each site, and were impractical.

Soil Factors
Four measures of soil conditions were recorded in situ. Soil depth and texture were recorded by digging a vertical sided pit 50cm deep. Sites with soil extending below this depth were recorded as 50cm (the approximate permafrost depth; deeper strata are inaccessible to plant roots and are therefore irrelevant). 5kg of soil from the pit was then passed through a 1cm sieve to extract stones; these stones were weighed on a spring balance, and the stone content of the soil expressed as percentage by mass. Differences in the water contents of the soil were neglected for this measurement.

Preliminary surveys revealed the maximum root density to occur at a depth of 10cm. Therefore, measurements of soil water content and pH were recorded at this depth, by vertically inserting the probe tip a fixed distance. Both quantities were measured by hand-held Fisons meters; water readings were converted from meter scale units to percentage by mass by the construction of a calibration curve. (5kg of soil was air dried and passed through a 2mm sieve. Known masses of water were added, the soil was mixed thoroughly, and the meter reading recorded as the average of 5 replicates.) Soil water content at the site was also recorded as the mean of 5 replicates, preliminary studies having revealed that this was the minimum number required to give no significant difference between the sample mean and the actual water content within the quadrat (as measured by the mean of 25 replicates.)

In addition, soil samples from each site were preserved for laboratory analysis (see later.)

Samples taken from the site
Samples of roots of Poa arctica and Saxifraga oppositifolia were taken at each site that these species were present. 1cm lengths of roots were taken from a minimum of five plants, where possible, and preserved in 95% methanol in 10ml screwtop perspex vials. It was initially planned to preserve samples in formyl acetic acid, but this was impounded by Norwegian customs and had to be replaced by meths bought locally.

In addition, soil samples were taken at each site. Surrounding vegetation was cleared away, and a standard metal can pressed into the exposed soil. This produced a standard core of approximately 400g, taking soil to a depth of 10cm. These soil cores were air dried, passed through a 2mm sieve, and 100g portions were preserved in ziplock plastic bags. Laboratory analysis of samples.

Laboratory analysis of the samples took place between November 14^th and December 3^rd at Cambridge University Plant Sciences department.

Soil samples were analysed for pH, exchangeable nitrogen and phosphorus, loss on ignition and for mycorrhizal fungal spore content. pH was measured with a W.G. Pye & co. meter in 2g samples made up into a slurry with 6g de-ionised water. Values produced correlated well to those measured in situ (Spearmann's rank correlation coefficient = 0.94), but were of greater precision. Exchangeable nitrogen and phosphorus were measured from soil solution extracts using a Chemlab continuous flow colorimeter. The meter was calibrated by passing 0.0, 0.2, 0.4, 0.6, 0.8 and 1.0 ppm standard solutions through it every 20 samples. In addition, standard soil (subsoil from Madingley wood, nr. Cambridge) samples were analysed to verify the accuracy of the analysis, and experimental values found to agree with the accepted values to within 0.2 ppm. Loss on ignition (soil organic matter) was measured from 5g soil samples placed in a 400^oC muffle furnace for 8 hours. Losses were expressed as percentages of initial dry mass.

Spore counts were produced from 5g samples of soil, and extracted by the wet sieving method of Nicolson. Spores were extracted from the soil fraction retained between 75 and 250mm sieves by resuspension in water in a petri dish, and were examined under x25 magnification with a Nikon dissecting microscope. No degraded or corroded spores were included in the count. Root samples were cleared, stained and mounted by the method of Grace and Stribley. 5 separate 0.5cm root segments from each sample were scanned under x40 magnification, and the percentage of root length infected by quantified by use of an eyepiece graticule. Mean infection densities were calculated from this data.

RESULTS
The most important result of the study was that of the ordination analysis. This showed that the strongest correlation between the major axis of variation and the environmental factors measured was with altitude. This was expected; there was a clear series of changes in habitat with altitude, from the dunes by the lake to the scree slopes above, and an associated series of changes in vegetation. However, altitude is an 'indirect' distribution factor; it does not affect plants in any meaningful way. Instead, the altitudinal trend reflects a number of factors directly affecting plants which also vary altitudinally. So what are these direct factors? Although many conventional causes of distribution, such as soil nutrients and light, are important, the extent of development of the mycorrhiza at the site is even more important still. The probability that this result occurs by chance is less than one in one thousand; highly statistically significant. Therefore, we may conclude that one of the important causes of the pattern of vegetation at a site, if not the most important, is the abundance of its mycorrhizal fungi.

Further confirmation of the importance of mycorrhiza in community pattern may be obtained by plotting the positions of plant species in ordination phase space; the position of the species on the graph represents the type of distribution pattern the species exhibits.

It was seen that the species with different types of mycorrhiza cluster in different parts of phase space. Therefore, species with different mycorrhizal types have different distribution patterns; mycorrhiza determine where a species occurs within the site.

However, the mechanism by which this occurs is less obvious. We might expect simply that sites with greater densities of mycorrhiza will favour plants which are infected by this type of fungus. In fact, this is not the case. The percentage cover of a species plotted against the abundance of its mycorrhizal fungus at the site shows no relationship between the two quantities.

An interesting conclusion of these analyses is that, although the presence of mycorrhiza benefits the plant, increasing the intensity of the infection may actually decrease the success of the species. This suggests that the benefits of the symbiosis do not increase with infection intensity (become saturated at relatively low intensities), whereas the costs of symbiosis (in terms of resources diverted to the fungus) continue to increase with intensity. Thus, the primary benefit to the host lies in being connected to the mycorrhizal network and not in the extent of the connections. This idea is supported by a plot of root infection density against the spore count in the soil, a measure of the fungal innoculum potential at the site.

Clearly, the relationship between the infection of the roots of a species (& benefits accruing to the species) is not linearly related to the availability of innoculum in the soil, but reaches a maximum at a relatively low innoculum potential.

DISCUSSION
The principal conclusion of the study is that mycorrhiza are an important cause of species distribution and plant community structure; different levels of mycorrhizal development produce different patterns of species abundance. However, the mechanism by which this occurs is less clear. Four possibilities may be distinguished:

1. Mycorrhiza foster individual species directly, through the symbiosis discussed in the introduction.
2. Mycorrhiza retard individual species directly, by secreting specific toxins, or invading their tissues.
3. Mycorrhiza affect species indirectly by mediating their responses to environmental stresses.
4. Mycorrhiza affect species indirectly by mediating the interactions between plants (primarily competitive).

Hence, a simple plot of species abundance against mycorrhizal density shows no trend because mycorrhiza are not important at every site. Even where they are important, the species for which they are important, and the way in which they act will vary. It is the ubiquity of these interactions that produce the effects on community structure; the effects shown at the level of the community are the summation of many such effects at the level of the species.

Therefore, to try to investigate the way in which mycorrhiza determine community structure by looking at their effects on individual species may be rather like looking at a jigsaw in terms of its pieces; the importance of the pieces is not intrinsic, but lies in the way in which they fit together. The most appropriate way in which to look at the jigsaw is to see the whole pattern of the interlocked pieces; the most appropriate way to look at the role of the mycorrhiza in the community is to look at all the species together, not in isolation. To look at a single species is like looking at a single jigsaw piece; no sense of its relationship to the wider picture is gained and therefore its true importance is lost.

However, it should be emphasised that these results are tentative, being based on a single study of an extremely simple ecosystem. Therefore, further replicate studies will be required, preferably in a variety of locations and ecosystems in order to confirm this finding. The conclusions drawn on the mechanism by which mycorrhiza act on community structure are also highly speculative, and based on correlative data. Therefore, logical extensions of the study would be either to investigate the consequences of artificially manipulating the mycorrhizal infection density in a natural ecosystem or to investigate the effect of infection on plant-plant interactions in a controlled, artificial system.

The theoretical importance of this study is two-fold. Firstly, it indicates that mycorrhiza are an important, and neglected, cause of plant distribution. This has important ramifications for ecological modelling, since it implies that previous models of the distribution of species and the response of species to environmental changes are incomplete. This view also lends support to Harley's model of the role of mycorrhiza in communities (Harley, 1982). Here, mycorrhiza were viewed as extensive networks of hyphae linking many individuals of many different species (such interspecific linkages are known to occur in the field). Thus the interactions between the roots of plants (below-ground competition) are not direct, but primarily due to the differential interactions between the two plants and their common mycorrhiza. In this way mycorrhiza act as a unifying or linking force in the community, somewhat analogous to the wiring between electrical appliances in a house. Secondly, and more generally, it suggests that the interactions between species, including those between taxonomically dissimilar organisms such as fungi and plants, are more important in determining species distribution and community structure than is commonly supposed. Indeed, they may even be more important than the physical environmental factors that are commonly cited. This leads us towards a new view of communities, in which the interactions between component species and not merely the species themselves are important. Therefore, the traditional methodology of studying communities by breaking them down into small parts or mechanisms is invalid, or at least inappropriate; communities should be studied at the level of the entire system, since the processes which structure them are not observable at the reductionist level.

A fuller experimental report has been submitted as a third-year dissertation for the Natural Sciences Tripos (Plant Sciences) at Cambridge University, and will hopefully be published in the Journal of Ecology in 1994.

This view, that even forces which have a profound effect in structuring communities may be localised in space & time and may apply only to a part of the community, is akin to the modern view of the role of intraspecific competition in communities.