Dealing with AFLP genotyping errors to reveal genetic structure in Plukenetia volubilis (Euphorbiaceae) in the Peruvian Amazon

Abstract

An analysis of the population structure and genetic diversity for any organism often depends on one or more molecular marker techniques. Nonetheless, these techniques are not absolutely reliable because of various sources of errors arising during the genotyping process. Thus, a complex analysis of genotyping error was carried out with the AFLP method in 169 samples of the oil seed plant Plukenetia volubilis L. from small isolated subpopulations in the Peruvian Amazon. Samples were collected in nine localities from the region of San Martin. Analysis was done in eight datasets with a genotyping error from 0 to 5%. Using eleven primer combinations, 102 to 275 markers were obtained according to the dataset. It was found that it is only possible to obtain the most reliable and robust results through a multiple-level filtering process. Genotyping error and software set up influence both the estimation of population structure and genetic diversity, where in our case population number (K) varied between 2-9 depending on the dataset and statistical method used. Surprisingly, discrepancies in K number were caused more by statistical approaches than by genotyping errors themselves. However, for estimation of genetic diversity, the degree of genotyping error was critical because descriptive parameters (He, FST, PLP 5%) varied substantially (by at least 25%). Due to low gene flow, P. volubilis mostly consists of small isolated subpopulations (ΦPT = 0.252-0.323) with some degree of admixture given by socio-economic connectivity among the sites; a direct link between the genetic and geographic distances was not confirmed. The study illustrates the successful application of AFLP to infer genetic structure in non-model plants.

Fig 1. Maps showing sampling locations of P. volubilis in the Peruvian Amazon.

Inset: Map A shows the San Martin region with the positions of P. volubilis sampling in detail. Map B shows the localization of San Martin within the whole of Peru.

Fig 2. Workflow.

Fig 3. Technical box.

Fig 4. Result of MDS analysis in the form of a 2D projection onto the plane for the “RawGeno” datasets.

This graphic projection represents individual samples from Dos de Mayo (2DM) as one cluster and the second cluster consists of all remaining samples from Aguas de Oro (ADO), Aucaloma (AUC), Chumbaquihui (CHU), Mishquiyacu (MIS), Pacchilla (PAC), Pucallpa (PUC), Ramón Castillo (RAC) and Santa Cruz (SCR).

Fig 5. Result of MDS analysis in the form of a 2D projection onto the plane for the “Error” datasets.

This graphic projection represents individual samples from Dos de Mayo (2DM) as one cluster and the second cluster consists of all remaining samples from Aguas de Oro (ADO), Aucaloma (AUC), Chumbaquihui (CHU), Mishquiyacu (MIS), Pacchilla (PAC), Pucallpa (PUC), Ramón Castillo (RAC) and Santa Cruz (SCR).

Fig 6

Mean ± S.D. (red vertical line) Ln(P)D value over 10 replicated runs for each estimated K = 1–9 on the left part of the figure (a) in the case of “standard” analysis. The chosen K is indicated by * for better clarity. The order of estimation Ln(P)D value for each dataset is equal to the dataset order on the side of graph bars. Please note the different scales of the Ln(P)D axes. The right part of the figure (b) shows a graph of each individual within the appropriate subpopulation indicated by the shortened name. The segmentation of vertical bars by different colors represents the estimated membership of an individual in K inferred clusters. There are two results for each dataset whenever K was estimated differently for the “standard” and “hierarchical” types of STRUCTURE analysis.