Isozyme Allelic Frequencies Related to Selection

Significant correlations between allelic frequencies and environmental variables in a number of insect species have been demonstrated by multivariate techniques. Since many environmental variables show a strong relationship to geographic location and since gene flow between populations can also produce patterns of gene frequencies which are related to the geographic location, both selection and gene-flow hypotheses are consistent with the observed correlations. The genetic variables can be corrected for geographic location and so for linear gene-flow patterns. If, after correction, the genetic variables still show significant correlations with similarly corrected environmental variables, then these correlations are consistent with hypotheses of selection but not of gene flow. The data of JOHNSON and SCHAFFER (1973) have been reanalyzed using the method of canonical correlation after correction for geographical location by means of multiple regression. Five of the nine loci studied exhibit significant canonical correlations. These results, under the assumption of linear gene flow, support hypotheses of selective action of environmental variables in the genotype-environment relationships observed. HE relationships between isozyme (allozyme) allelic frequencies in natural populations and various measurements of the environments in which these populations exist have been studied in a number of insect species. JOHNSON et al. ( 1969) found statistically significant relationships in the harvester ant, Pogonrrzyrmex barbatus. KOJIMA et al. (1972) studied Drosophila pavani, ROCKWOODSLUSS, JOHNSTON and HEED (1973) studied D. pachea, TOMASZEWSKI, SCHAFFER and JOHNSON (1973) studied P. badius, and JOHNSON and SCHAFFER (1973) studied D. melanogaster. In all of these cases, statistically significant relationships were demonstrated between allelic frequency variation and some characteristics of the environment. The environmental characteristics include the temperature, precipitation and humidity records maintained by the Weather Bureau. ROCKWOOD-SLUSS, JOHNSTON and HEED (1973) also included the concentration of several chemical compounds in the senita cactus which is the food source for D. pachea. ' Paper Number 4173 of the Journal Series of the North Carolina State Agricultural Experiment Station, Raleigh, N.C. This investigation was supported by NIH Research Grant Number GM 11546 from the National Institute of General Medical Sciences and by contract number AT-(W-1)3980 from the U.S. Atomic Energy Commission. Genetics 77: 163-168 May 1974. 164 H. E. SCHAFFER -4ND F. M . J O H N S O N The statistical methodology used in these studies, except for in that of KOJIMA et al. (1972), has primarily involved multivariate methods such as principal components and canonical correlation; TAYLOR and MITTON (1972, 1974) have used factor analysis. All of these methods are used to synthesize new variables from the original variables. The new variables can be considered to be patterns which are identified in the original data. These patterns are sought in the data to satisfy such criteria as maximum variance or correlation, depending on which multivariate method is used. The end result in each study has been the identification of allelic frequency (genetic) patterns and environmental patterns which exhibit a statistically significant correlation. Most of the environmental patterm identified in the studies mentioned above have shown a relationship with the geographical locations at which the populations were collected and the environmental measurements taken. It is natural that such environmental characteristics as elevation, temperature and precipitation show strong relationships with geographical location. Thus when patterns of these variables show correlations with genetic patterns, hypotheses based on selective effects of the environment become confounded with those based on non-selective location-dependent effects, e.g., gene flow as a result of drift and migration. This confounding of the two categories of hypotheses has been pointed by SCHAFFER and JOHNSON (1 973). However, when the environmental and genetic patterns are not completely correlated with geographical location, it is possible to partially differentiate between effects of the two hypotheses. The results of JOHNSON and SCHAFFER (1973) will be reanalyzed in such a manner. MODEL A N D ANALYSIS A cline for an allelic frequency may arise as the result of a wave of migration, in which the frequency of migrants carrying a new allele gradually diminishes over distance, or from the meeting of two populations with different frequencies of an allele. Considering a two-dimensional geographical area instead of a transect and letting the frequency of the allele represent the third dimension, such a cline will appear as a surface. In the simplest case, the allelic frequency will be proportional to the distance involved and so the allelic frequency surface will be a plane. Subsequent instances of gene flow for the same allele may alter the orientation of the plane but will leave it as a plane. If more than two alleles are present, then additional allele will add one more dimension and the graph can no longer be directly visualized, as it requires more than three dimensions for its representation. However, the surface described in the higher dimensional space, termed a flat, is analogous to the plane which was visualized in three dimensions (KENDALL 1961). Clines in gene frequencies can also be produced as the result of the intergradation of two populations. More than two populations may also be involved. Such intergradation can result in a cline which tends to be a linear combination of the genes of the two populations with coefficients proportional to the distance from the ends of the overlapping region. Such a cline should be, to a first approximation, a “flat”. SELECTION AND GENE-FLOW HYPOTHESIS 165 If the observed allelic frequencies at a locus are plotted as additional dimensions versus the geographical locations at which the collections were made, a higher dimensional representation will result. To the extent that the points representing the allelic frequencies fit a flat, the differences in allelic frequencies between locations could have resulted from the type of gene flow described above. Many environmental factors also approximate a flat when plotted against the geographical coordinates. Thus selective adaptation of allelic frequencies to these factors would also tend to produce allelic frequency points which are consistent with a flat. In this way, two possible causative agencies are confounded. Environmental variables are also observed to be in patterns related to the geographical location. The strongest such relationship in the environmental data presented by JOHNSON and SCHAFFER (1973) was for average annual temperature, which has 98.5% of its variation linearly related to geographical location. The weakest was average noon relative humidity with 1.5 % of its variation related to location. In each case this percentage was the multiple correlation coefficient ( R 2 ) resulting from a multiple regression of the environmental variable on the independent variables of latitude and longitude. In all of these analyses the geographical area is being treated as a two-dimensional area with latitude and longitude comprising a rectangular coordinate system. To the extent that both the environmental variables and the possible results of gene migration show similar patterns, either of them could explain equally well the appearance of such patterns in the gene frequencies. The presence of these patterns can be seen by the size of the R2 values given in Table 1. The re-