peerj_json_files/PeerJ_Json_114.json

{
    "v1_Abstract": "In microbial ecology studies, the most commonly used ways of investigating alpha (withinsample) diversity are either to apply non-phylogenetic measures such as Simpson's index to Operational Taxonomic Unit (OTU) groupings, or to use classical phylogenetic diversity (PD), which is not abundance-weighted. Although alpha diversity measures that use abundance information in a phylogenetic framework do exist, but are not widely used within the microbial ecology community. The performance of abundance-weighted phylogenetic diversity measures compared to classical discrete measures has not been explored, and the behavior of these measures under rarefaction (sub-sampling) is not yet clear. In this paper we compare the ability of various alpha diversity measures to distinguish between different community states in the human microbiome for three different data sets. We also present and compare a novel one-parameter family of alpha diversity measures, \\ (\\operatorname{BWPD}_\\theta\\), that interpolates between classical phylogenetic diversity (PD) and an abundance-weighted extension of PD. Additionally, we examine the sensitivity of these phylogenetic diversity measures to sampling, via computational experiments and by deriving a closed form solution for the expectation of phylogenetic quadratic entropy under re-sampling. In all three of the datasets considered, an abundance-weighted measure is the best differentiator between community states. OTU-based measures, on the other hand, are less effective in distinguishing community types. In addition, abundance-weighted phylogenetic diversity measures are less sensitive to differing sampling intensity than their unweighted counterparts. Based on these results we encourage the use of abundanceweighted phylogenetic diversity measures, especially for cases such as microbial ecology where species delimitation is difficult.",
    "v2_Abstract": "In microbial ecology studies, the most commonly used ways of investigating alpha (within-sample) diversity are either to apply count-only measures such as Simpson's index to Operational Taxonomic Unit (OTU) groupings, or to use classical phylogenetic diversity (PD), which is not abundance-weighted. Although alpha diversity measures that use abundance information in a phylogenetic framework do exist, but are not widely used within the microbial ecology community. The performance of abundance-weighted phylogenetic diversity measures compared to classical discrete measures has not been explored, and the behavior of these measures under rarefaction (sub-sampling) is not yet clear. In this paper we compare the ability of various alpha diversity measures to distinguish between different community states in the human microbiome for three different data sets. We also present and compare a novel one-parameter family of alpha diversity measures, BWPD \u03b8 \\operatorname{BWPD}_\\theta , that interpolates between classical phylogenetic diversity (PD) and an abundance-weighted extension of PD. Additionally, we examine the sensitivity of these phylogenetic diversity measures to sampling, via computational experiments and by deriving a closed form solution for the expectation of phylogenetic quadratic entropy under re-sampling. In all three of the datasets considered, an abundance-weighted measure is the best differentiator between community states. OTU-based measures, on the other hand, are less effective in distinguishing community types. In addition, abundance-weighted phylogenetic diversity measures are less sensitive to differing sampling intensity than their unweighted counterparts. Based on these results we encourage the use of abundance-weighted phylogenetic diversity measures, especially for cases such as microbial ecology where species delimitation is difficult.",
    "v1_text": "in microbial ecology studies, the most commonly used ways of investigating alpha (within- : sample) diversity are either to apply non-phylogenetic measures such as Simpson's index to Operational Taxonomic Unit (OTU) groupings, or to use classical phylogenetic diversity (PD), which is not abundance-weighted. Although alpha diversity measures that use abundance information in a phylogenetic framework do exist, but are not widely used within the microbial ecology community. The performance of abundance-weighted phylogenetic diversity measures compared to classical discrete measures has not been explored, and the behavior of these measures under rarefaction (sub-sampling) is not yet clear. In this paper we compare the ability of various alpha diversity measures to distinguish between different community states in the human microbiome for three different data sets. We also present and compare a novel one-parameter family of alpha diversity measures, \\ (\\operatorname{BWPD}_\\theta\\), that interpolates between classical phylogenetic diversity (PD) and an abundance-weighted extension of PD. Additionally, we examine the sensitivity of these phylogenetic diversity measures to sampling, via computational experiments and by deriving a closed form solution for the expectation of phylogenetic quadratic entropy under re-sampling. In all three of the datasets considered, an abundance-weighted measure is the best differentiator between community states. OTU-based measures, on the other hand, are less effective in distinguishing community types. In addition, abundance-weighted phylogenetic diversity measures are less sensitive to differing sampling intensity than their unweighted counterparts. Based on these results we encourage the use of abundance- weighted phylogenetic diversity measures, especially for cases such as microbial ecology where species delimitation is difficult. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Connor O. McCoy and Frederick A. Matsen IV\u2217 Fred Hutchinson Cancer Research Center 1100 Fairvew Ave. N Seattle, WA 98109 \u2217Corresponding author: matsen@fhcrc.org 1 PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 2 1. INTRODUCTION It is now well accepted that incorporating phylogenetic information into alpha (single-sample) and beta (between-sample) diversity measures can be useful in a variety of ecological contexts. Phylogenetic equivalents of all of major alpha diversity measures have been developed (Table 1). Starting with Faith\u2019s original definition of phylogenetic diversity (Faith, 1992), which generalizes species count, there are now phylogenetic generalizations of the Simpson index to Rao\u2019s quadratic entropy (Rao, 1982; Warwick and Clarke, 1995), the Shannon index to phylogenetic entropy (Allen et al., 2009), and the Hill numbers to qD(T) (Chao et al., 2010). Phylogenetic diversity itself has been extended to incorporate taxon counts (Barker, 2002) and proportional abundance (Vellend et al., 2011). There have also been abundance-weighted measures that explicitly measure phylogenetic community structure (Fine and Kembel, 2011), or an \u201ceffective number of species\u201d (Chao et al., 2010). Many diversity measures can be tidily expressed in the framework of Leinster and Cobbold (2012), although the expression of phylogenetic diversity measures for non-ultrametric trees is complex. In this paper we use three example human microbiome datasets to demonstrate the utility of abundance-weighted phylogenetic diversity measures. We also introduce a one-parameter family interpolating between classical PD and an abundance-weighted generalization. We call the parameter \u03b8 and denote the one-parameter family BWPD\u03b8; BWPD0 is classical PD, whereas BWPD1 is balanceweighted phylogenetic diversity, effectively PDaw of Vellend et al. (2011). Intermediate values of \u03b8 allow a partially-abundance-weighted compromise. Such a compromise has recently been shown to be useful for measuring beta diversity, with the introduction of a oneparameter family of \u201cgeneralized UniFrac\u201d measures (Chen et al., 2012). We use the name Balance Weighted Phylogenetic Diversity as described below because there are a variety of abundance weighted phylogenetic diversity measures. We compare the behavior of PD measures, including BWPD\u03b8, under various levels of sampling using theory and example data sets. 2. MATERIALS AND METHODS 2.1. Datasets. We apply the methods described below to three previously described 16S rRNA surveys of the human microbiome. The PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 3 first two datasets are composed of samples from \u201cnormal\u201d and dysbiotic microbial communities, where previous studies have associated changes in diversity with changes in health. The third dataset investigates the changes of the skin microbiome through time. 2.1.1. Bacterial vaginosis. First, we reanalyze a pyrosequencing dataset describing bacterial communities from women being monitored in a sexually transmitted disease clinic for bacterial vaginosis (BV). BV has previously been shown to be associated with increased community diversity (Fredricks et al., 2005). For this study, swabs were taken from 242 women from the Public Health, Seattle and King County Sexually Transmitted Diseases Clinic between September 2006 and June 2010 of which 220 samples resulted in enough material to analyze (Srinivasan et al., 2012). Selection of reference sequences and sequence preprocessing were performed using the methods described in (Srinivasan et al., 2012). 452,358 reads passed quality filtering, with a median of 1,779 reads per sample (range: 523\u20132,366). 2.1.2. Oral periodontitis. We also utilize sequence data from a study of subgingival communities in 29 subjects with periodontitis, along with an equal number of healthy controls (Griffen et al., 2011a). The publication analyzing this dataset showed increased community diversity in samples from dysbiotic patients compared to healthy controls. Raw sequences were filtered, retaining only those reads with: a mean quality score of at least 25, no ambiguous bases, at least 150 base pairs in length, and an exact match to the sequencing primer and barcode. A total of 759,423 reads passed quality filtering, with a median of 8,320 reads per sample (range: 4,096\u201314,319). As the phylogenetic placement method used below to calculate our measures requires a reference tree and alignment, we created a tree with FastTree 2.1.4 (Price et al., 2010) using the alignment and accompanying taxonomic annotation from the curated CORE database of oral microbiota (Griffen et al., 2011b). 2.1.3. Skin microbiome through time. Our third data set is a study of skin microbial diversity through adolescence (Oh et al., 2012). Aligned sequences were obtained courtesy of the authors, although sequence data is available under the accession numbers [GQ000001] to [GQ116391] and can be accessed through BioProject ID 46333. A total of 90,142 Sanger sequences were available, with a median of 693 sequences per sample (range: 317\u20132884). PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 4 2.2. Balance-weighted phylogenetic diversity. In this section we introduce BWPD\u03b8, our one-parameter family interpolating between classical PD and fully balance-weighted phylogenetic diversity. We will primarily consider so-called unrooted (Pardi and Goldman, 2007) phylogenetic diversity, which does not necessarily include the root. The case of rooted phylogenetic diversity can be calculated in a similar though simpler way as described below. Although we will primarily be working in an unrooted sense, it will be useful to use terminology that corresponds to the rooted case. For this reason, if the tree is not already rooted, assume an arbitrary root has been chosen; let the proximal side of a given edge be the side that contains the root and distal be the other. We will describe BWPD\u03b8 in terms of a phylogenetic tree T with leaves L, and a contingency table describing the number of observations of the organisms at the leaves in various samples. The contingency table has rows labeled with the leaves of T , and columns labeled by samples. In microbial ecology this is frequently known as an OTU table. The entry corresponding to a given leaf and a given sample is the number of times that leaf was observed in that sample. The classical (unrooted) phylogenetic diversity of a given sample in this context is the total branch length of the tree subtended by the leaves in that sample. The path to generalizing PD is to note that this can be expressed as a sum of branch lengths multiplied by a step function. Let f(x) be the function that is one for x > 0 and zero otherwise. Let g(x) = min(f(x), f(1\u2212x)) and Ds(i) be the fraction of reads in sample s that are in leaves on the distal side of edge i. Phylogenetic diversity can be then expressed as (1) PDu(s) = \ufffd i \ufffdi g(Ds(i)) That is, the sum of edge lengths in T which have reads from s on both the distal and proximal side. Note that the step function g is the limit of a one-parameter family of functions (Fig. 1). Indeed, defining (2) g\u03b8(x) = [2min(x, 1\u2212 x)]\u03b8 , g is the pointwise limit of the g\u03b8 on the closed unit interval as \u03b8 goes to zero. Thus our one-parameter generalization is (3) BWPD\u03b8(s) = \ufffd i \ufffdi g\u03b8(Ds(i)). PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 5 Note that when \u03b8 = 0 this is PD and when \u03b8 = 1 this is an abundanceweighted version of PD equivalent to executing the \u0394nPD recipe of Barker (2002) up to a multiplicative factor. The rooted equivalent of (3) is (4) RBWPD\u03b8(s) = \ufffd i \ufffdi (Ds(i)) \u03b8, which interpolates between rooted PD and an abundance-weighted version. Vellend et al. (2011) describe a measure, PDaw, which is equal to RBWPD1 multiplied by the total number of branches in T . We call BWPD1 balance-weighted phylogenetic diversity because it weights edges according to the balance of read fractions on either side of an edge\u2013 edges with even amount of mass on either side are up-weighted, while edges with an uneven balance of mass are down-weighted. Indeed, if |x \u2212 (1 \u2212 x)| is thought of as the imbalance of read fraction on either side of an edge, then 1\u2212 |x\u2212 (1\u2212x)| is a measure of balance; note that on the unit interval, 2min(x, 1\u2212 x) = 1\u2212 |x\u2212 (1\u2212 x)|. Because a small x or an x close to 1 gives a small coefficient in the summation, small collections of reads or small perturbations of the read distribution will not change the value of BWPD1 appreciably. 2.3. Calculation of PD measures in example applications. Reads from the vaginal and oral studies were placed on a tree created from a curated set of taxonomically annotated reference sequences. As phylogenetic entropy and qD(T) operate on a rooted phylogeny, reference trees were assigned a root taxonomically (Matsen and Gallagher, 2012) meaning that a root was found that best separated highlevel taxonomic groupings. pplacer was run in posterior probability mode (using the -p and --informative-prior flags), which defines an informative prior for pendant branch lengths with a mean derived from the average distances from the edge in question to the leaves of the tree. The resulting set of placements were classified at the family rank using a hybrid classifier implemented in the guppy tool from the pplacer suite. The hybrid classifier assigns taxonomic annotations to sequences using the combination of a na\u0131\u0308ve Bayes classifier (Wang et al., 2007) with a phylogenetic classifier (Matsen et al., unpublished results). Any reads that could not be confidently classified to the family rank were not used in measures based on classification. Full-length 16S sequences were available for the skin data, and so a more traditional tree-building approach was used. Representative PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 6 OTUs were chosen for each site by clustering at 97% identity using USEARCH 5.1 (Edgar, 2010), with trees built on OTU centroids using FastTree (Price et al., 2010). To conform with methods used in Oh et al. (2012), the na\u0131\u0308ve Bayes classifier (Wang et al., 2007) was used to infer genus-level classifications to taxonomically root the tree; in our case we used the RDP classifier v2.5. The contingency (OTU) tables generated by clustering were made available to our tools via the BIOM (McDonald et al., 2012) format. PDu (unrooted PD), phylogenetic quadratic entropy (Rao, 1982), phylogenetic entropy (Allen et al., 2009), and qD(T) (Chao et al., 2010) were implemented for phylogenetic placements in the freelyavailable pplacer suite of tools (Matsen et al., 2010) (http://matsen. fhcrc.org/pplacer) in the subcommand guppy fpd. Prior to diversity estimation, either phylogenetic placements were rarefied to the read count of the specimen in the dataset with the fewest sequences using guppy rarefy, or the corresponding rarefaction on fulllength sequences was performed with the QIIME (Caporaso et al., 2010) rarefaction tool single rarefaction.py. The mean value of each statistic over 100 such rarefactions was used for analysis. Discrete measures of alpha diversity and richness were calculated on contingency tables obtained from clustering and taxonomic classification. Sequences were clustered into Operational Taxonomic Units (OTUs) at a 97% identity threshold using USEARCH 5.1 (Edgar, 2010). Similar results were observed when clustering at 95% identity (results not shown). OTU counts and family-level taxon counts were then rarefied as above in R 3.0.1 (R Development Core Team, 2012) using the vegan package (Oksanen et al., 2012). We obtained values for the Simpson (1949) and Shannon (1948) diversity indices, as well as the Chao1 (Chao, 1984) and ACE (Chao and Lee, 1992) measures of species richness using vegan functions diversity and estimateR. 2.4. Comparative analysis of alpha diversity measures. To investigate the relation between various measures of alpha diversity, we calculated Pearson\u2019s r between all pairs of measures using the function rcorr from the R package Hmisc (Harrell Jr., 2012). We then performed hierarchical clustering with the R function hclust, using d = 1\u2212 r as the distance between two measures. Association of each measure with clinical criteria for the first two data sets was evaluated by examining the accuracy of a logistic regression using the measure as the sole predictor of whether the sample came from a \u201cnormal\u201d or dysbiotic subject. In the vaginal dataset, we assessed each measure\u2019s ability to predict whether a sample was PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 7 from a subject positive for BV by Amsel\u2019s criteria, a clinical diagnostic method (Amsel et al., 1983). In the oral dataset, we assessed each measure\u2019s ability to predict whether a sample was from a healthy control, or a subject with periodontitis. Accuracy in predicting sample community state was assessed by leave-one-out cross-validation using the R package boot (Davison and Hinkley, 1997; Canty and Ripley, 2012). For the vaginal dataset, we also calculated R2 values using each measure individually as a predictor for sample Nugent score in a linear regression. The Nugent score provides a diagnostic score for BV which ranges from 0 (BV-negative) to 10 (BV-positive) based on presence and absence of bacterial morphotypes as viewed under a microscope (Nugent et al., 1991). We calculated p-values to compare within- and between-stratification variability using R\u2019s built-in t.test function for the vaginal data, which had a binary stratification, and the aov function for the oral and skin data sets. The vaginal dataset data was stratified by Amsel\u2019s criterion, the oral dataset by condition and sampling site, and the skin microbiome dataset by Tanner scale of physical development (Oh et al., 2012). Note that we are not presenting these uncorrected pvalues as evidence that there is an interesting relationship between the microbiome and a given stratification, but rather are using pvalues as a way of measuring within-stratum heterogeneity compared to between-stratum heterogeneity for the various measures. 3. RESULTS 3.1. Application to the human microbiome. 3.1.1. Vaginal microbiome. Like Srinivasan et al. (2012) and many others in the field, we observe greater diversity in BV positive specimens using a variety of diversity and richness measures (Fig. S1). In particular, this is true for BWPD\u03b8 for a variety of values of \u03b8 (Fig. S2). In the vaginal data, phylogenetic measures of alpha diversity have better cross-validation accuracy for the Amsel classification and better correlation with the Nugent score than discrete OTU-based measures (Table 2). All measures were somewhat accurate in identifying community state, with even the worst performers classifying almost 70% of samples correctly. BWPD0.25, BWPD0.5, PDu, and phylogenetic entropy perform well predicting BV status. Correlation with Nugent score varies from 0.19 using Simpson (OTU) to 0.74 using PDu. OTU-based measures rank in the bottom half of the measures tested, and below all phylogenetic measures. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 8 In the hierarchical clustering of alpha measures on the vaginal data set, phylogenetic methods are separated from OTU-based methods (Fig. 2). BWPD\u03b8 is similar to different extant phylogenetic alpha diversity measures for different \u03b8. The Simpson and Shannon diversity measures cluster together, as do the ACE and Chao1 richness measures. Fig. 3 shows values of BWPD\u03b8 calculated before (x-axis) and after (y-axis) a single rarefaction to 523 sequences per sample. Samples for which the BWPD\u03b8 value changes little lie close to the blue line, which shows the case of no difference between original and rarefied samples. Increasing \u03b8, which corresponds to increased use of abundance information, reduces the change in BWPD\u03b8 induced by rarefaction. Phylogenetic quadratic entropy and phylogenetic entropy both show behavior similar to BWPD1, with rarefaction introducing little effect. It might be possible to formalize a statement to this effect by computing the expectation of these alpha measures under rarefaction. However, computing the expectation for BWPD\u03b8 under rarefaction does not appear to be straightforward: the methods of Dremin (1994) might be applicable in this setting, however, even the integer moments of the hypergeometric distribution are complicated and the non-integer moments are bound to be very complex. We have, however, shown in the Appendix that the expectation of phylogenetic quadratic entropy under rarefaction to k sequences assigned to the tips of a phylogenetic tree is E[PQEk] = k \u2212 1 kn(n\u2212 1) \ufffd i \ufffdidi(n\u2212 di) where di is the number of sequences falling below edge i and \ufffdi is the length of edge i. This is almost identical to the unrarefied value of phylogenetic quadratic entropy, i.e. PQE = 1 n2 \ufffd i \ufffdidi(n\u2212 di). Thus it is not surprising to see that the expectation of PQE under rarefaction is very close to the original value (Fig. S3) for reasonably large k and n. 3.1.2. Oral microbiome. As previously observed by Griffen et al. (2011a), we find generally higher diversity in samples from dysbiotic patients (Fig. 4). We evaluated the ability of each alpha diversity measure to PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 9 predict whether a sample came from an individual with periodontitis, regardless of sample collection site, using the above methods. In the oral dataset, phylogenetic alpha diversity measures incorporating abundance gave the best predictions of community state (Table 3, Fig. 4). In contrast, classical phylogenetic diversity performed less well. These results were almost identical in terms of rank order after applying additional quality filtering steps to correct sequencing errors and remove potentially chimeric sequences (Table S1). OTU-based methods and phylogenetic methods are not as separated in a hierarchical clustering as for the vaginal dataset (Fig. S6). However, many of the same pairings are present in both clusterings: BWPD0.5 with phylogenetic entropy, BWPD1 with quadratic entropy, Simpson with Shannon, and ACE with Chao1. Interestingly, PDu, BWPD0.25, and the qD(T) measures all cluster with the discrete richness measures ACE and Chao1. Like the vaginal dataset, incorporating abundance information decreases the effect of rarefaction on BWPD\u03b8 values (Figs. S4, S5). 3.1.3. Skin microbiome. To further assess resolution and robustness of phylogenetic diversity measures, we considered skin microbiome data from a study by Oh et al. (2012). This study tracked the changes of the skin microbiome through \u201cTanner\u201d developmental stages of adolescence(Tanner and Whitehouse, 1976). Because there are five Tanner stages, and they do not have a monotonic relationship with skin microbiome diversity (Oh et al., 2012), we focused on ANOVA p-values to see if the diversity measurements had small within-stage heterogeneity compared to between-stage heterogeneity. To compare the ANOVA p-values associated with the diversity measurements across the various data sets, we ranked the p-value of the diversity measures from lowest to highest for each data set individually. We averaged these ranks to gain an overall measure of performance. The results again show phylogenetic measures generally performing better than OTU-based measures (Table 4). This finding holds true even after removing potential chimeras (Table S1). In this case, a light weighting or no weighting of phylogenetic diversity by abundance performed better than full abundance-weighting. 3.1.4. Applications summary. In all three of the data sets investigated, abundance-weighted phylogenetic diversity measures showed good performance to distinguish between community states: between \u201cnormal\u201d and dysbiotic samples in the oral and vaginal microbiomes, and between developmental stages in the skin microbiome. Notably, PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 10 the best distinguishing measure in each dataset was phylogenetic; in addition BWPD0.25 and BWPD0.5 were the only measures that were in the top four for all data sets. The result that partial abundance weighting performs well corresponds to analogous results for beta diversity, where an intermediate exponent for \u201cgeneralized UniFrac\u201d was the most powerful (Chen et al., 2012). 4. DISCUSSION Phylogenetic alpha diversity measures were more closely related to community state than were discrete measures based on OTU clustering for the data sets investigated here. This result is especially interesting given that the Simpson index, the Shannon index, or counting applied to OTU tables are very common ways of characterizing microbial diversity (Fierer et al., 2007; Grice et al., 2009; Hill et al., 2003; Dethlefsen and Relman, 2011). As also noted by Aagaard et al. (2012), we find that measurements of diversity using taxonomic classification can be useful in describing communities, and in fact perform much better than the same measurements of diversity applied to OTU counts; however, this approach requires a taxonomically well characterized environment. Our results can be viewed as an experimental confirmation of the notion that incorporating similarity between species is important to get sensible measures of diversity, which has been advocated by many, including most recently by Leinster and Cobbold (2012). We find that classical phylogenetic diversity is sensitive to sampling depth, underestimating the true value in small samples. Biases have also been described for diversity measures using OTU tables (Gihring et al., 2012). In contrast, we observe that some abundanceweighted phylogenetic measures are relatively robust to varying levels of sampling. These results did not appear to be the result of of sequencing issues. In principle, OTU methods could have performed badly because of error-prone and chimeric sequences inflating the number of OTUs. Although this is a real danger for OTU quantification, in this study its impact appears to be limited\u2013 similar results were obtained with the oral data after de-noising and chimera removal and the skin data (which used Sanger sequencing) after chimera removal. We note that on our data, non-phylogenetic measures applied to family level taxonomic groupings are generally more discriminating than the corresponding measures applied to OTUs. This may be PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 11 because our sequences are from the human microbiome, and taxonomic classification is especially well-developed in that setting; in particular the taxonomic names may already be defined in a way that corresponds to dysbiosis. Thus this particular difference may not continue to be true in a taxonomically less-well-characterized environment. As of the publication of this paper, no abundance-weighted phylogenetic alpha diversity measures are implemented in either mothur (Schloss et al., 2009) or QIIME (Caporaso et al., 2010), two of the most popular tools for analysis of microbial ecology data. Although the fact that abundance-weighted phylogenetic diversity measures performed very well for the three data sets investigated here does not imply that they are best in general, we suggest that abundanceweighted phylogenetic measures be given greater consideration for microbial ecology studies. For this to happen, implementations in commonly used microbial ecology software packages will be needed, in addition to our implementation and that of the picante R package (Kembel et al., 2010). 5. ACKNOWLEDGEMENTS The authors would like to thank Steven Kembel for encouragement and guidance, Steven N. Evans for probability consultation, and David Nipperess for an interesting dialog concerning phylogenetic diversity and rarefaction. The Segre lab at the NIH, in particular Sean Conlan and Julia Oh, were very generous and helpful with the skin data. This work would not have been possible without an ongoing collaboration with David Fredricks, Noah Hoffman, Martin Morgan, and Sujatha Srinivasan at the Fred Hutchinson Cancer Research Center. This work was supported in part by NIH R01 HG005966-01. REFERENCES K. Aagaard, K. Riehle, J. Ma, N. Segata, T.A. Mistretta, C. Coarfa, S. Raza, S. Rosenbaum, I. Van den Veyver, A. Milosavljevic, D. Gevers, C. Huttenhower, J. Petrosino, and J. Versalovic. A metagenomic approach to characterization of the vaginal microbiome signature in pregnancy. PLOS ONE, 7(6):e36466, 2012. B. Allen, M. Kon, and Y. Bar-Yam. A new phylogenetic diversity measure generalizing the Shannon index and its application to phyllostomid bats. The American Naturalist, 174(2):236\u2013243, 2009. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 12 R. Amsel, P.A. Totten, C.A. Spiegel, K. Chen, D. Eschenbach, and K.K. Holmes. Nonspecific vaginitis: diagnostic criteria and microbial and epidemiologic associations. The American Journal of Medicine, 74(1):14\u201322, 1983. G.M. Barker. Phylogenetic diversity: a quantitative framework for measurement of priority and achievement in biodiversity conservation. Biological Journal of the Linnean Society, 76(2):165\u2013194, 2002. A. Canty and B.D. Ripley. boot: Bootstrap R (S-Plus) Functions, 2012. R package version 1.3-5. J.G. Caporaso, J. Kuczynski, J. Stombaugh, K. Bittinger, F.D. Bushman, E.K. Costello, N. Fierer, A.G. Pen\u0303a, J.K. Goodrich, J.I. Gordon, et al. Qiime allows analysis of high-throughput community sequencing data. Nature Methods, 7(5):335\u2013336, 2010. A. Chao. Nonparametric estimation of the number of classes in a population. Scandinavian Journal of Statistics, 11(4):pp. 265\u2013270, 1984. A. Chao and S.M. Lee. Estimating the number of classes via sample coverage. Journal of the American Statistical Association, pages 210\u2013 217, 1992. A. Chao, C.H. Chiu, and L. Jost. Phylogenetic diversity measures based on Hill numbers. Philosophical Transactions of the Royal Society B: Biological Sciences, 365(1558):3599\u20133609, 2010. J. Chen, K. Bittinger, E.S. Charlson, C. Hoffmann, J. Lewis, G.D. Wu, R.G. Collman, F.D. Bushman, and H. Li. Associating microbiome composition with environmental covariates using generalized UniFrac distances. Bioinformatics, 28(16):2106\u20132113, 2012. A.C. Davison and D.V. Hinkley. Bootstrap Methods and Their Applications. Cambridge University Press, Cambridge, 1997. ISBN 0-521- 57391-2. L. Dethlefsen and D.A. Relman. Incomplete recovery and individualized responses of the human distal gut microbiota to repeated antibiotic perturbation. Proceedings of the National Academy of Sciences, 108(Supplement 1):4554, 2011. I.M. Dremin. Fractional moments of distributions. JETP letters c/c of Pis\u2019ma v zhurnal eksperimental\u2019noi teoreticheskoi fiziki, 59:585\u2013585, 1994. R.C. Edgar. Search and clustering orders of magnitude faster than BLAST. Bioinformatics, 26(19):2460\u20132461, 2010. D.P. Faith. Conservation evaluation and phylogenetic diversity. Biological Conservation, 61(1):1\u201310, 1992. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 13 N. Fierer, M. Breitbart, J. Nulton, P. Salamon, C. Lozupone, R. Jones, M. Robeson, R.A. Edwards, B. Felts, S. Rayhawk, et al. Metagenomic and small-subunit rRNA analyses reveal the genetic diversity of bacteria, archaea, fungi, and viruses in soil. Applied and Environmental Microbiology, 73(21):7059\u20137066, 2007. P.V.A. Fine and S.W. Kembel. Phylogenetic community structure and phylogenetic turnover across space and edaphic gradients in western amazonian tree communities. Ecography, 34(4):552\u2013565, 2011. D.N. Fredricks, T.L. Fiedler, and J.M. Marrazzo. Molecular identification of bacteria associated with bacterial vaginosis. New England Journal of Medicine, 353(18):1899\u20131911, 2005. T.M. Gihring, S.J. Green, and C.W. Schadt. Massively parallel rRNA gene sequencing exacerbates the potential for biased community diversity comparisons due to variable library sizes. Environmental Microbiology, 2012. E.A. Grice, H.H. Kong, S. Conlan, C.B. Deming, J. Davis, A.C. Young, G.G. Bouffard, R.W. Blakesley, P.R. Murray, E.D. Green, et al. Topographical and temporal diversity of the human skin microbiome. Science, 324(5931):1190\u20131192, 2009. A.L. Griffen, C.J. Beall, J.H. Campbell, N.D. Firestone, P.S. Kumar, Z.K. Yang, M. Podar, and E.J. Leys. Distinct and complex bacterial profiles in human periodontitis and health revealed by 16S pyrosequencing. The ISME Journal, 2011a. A.L. Griffen, C.J. Beall, N.D. Firestone, E.L. Gross, J.M. DiFranco, J.H. Hardman, B. Vriesendorp, R.A. Faust, D.A. Janies, and E.J. Leys. CORE: a phylogenetically-curated 16S rDNA database of the core oral microbiome. PLOS ONE, 6(4):e19051, 2011b. F.E. Harrell Jr. Hmisc: Harrell Miscellaneous, 2012. URL http: //CRAN.R-project.org/package=Hmisc. R package version 3.9-3. T.C.J. Hill, K.A. Walsh, J.A. Harris, and B.F. Moffett. Using ecological diversity measures with bacterial communities. FEMS Microbiology Ecology, 43(1):1\u201311, 2003. S.W. Kembel, P.D. Cowan, M.R. Helmus, W.K. Cornwell, H. Morlon, D.D. Ackerly, S.P. Blomberg, and C.O. Webb. Picante: R tools for integrating phylogenies and ecology. Bioinformatics, 26(11):1463\u2013 1464, 2010. Tom Leinster and Christina A Cobbold. Measuring diversity: the importance of species similarity. Ecology, 93(3):477\u2013489, 2012. F.A. Matsen and A. Gallagher. Reconciling taxonomy and phylogenetic inference: formalism and algorithms for describing discord and inferring taxonomic roots. Algorithms for Molecular Biology, 7 PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 14 (1):8, 2012. F.A. Matsen, R.B. Kodner, and E. Armbrust. pplacer: linear time maximum-likelihood and Bayesian phylogenetic placement of sequences onto a fixed reference tree. BMC Bioinformatics, 11(1):538, 2010. D. McDonald, J.C. Clemente, J. Kuczynski, J.R. Rideout, J. Stombaugh, D. Wendel, A. Wilke, S. Huse, J. Hufnagle, F. Meyer, et al. The Biological Observation Matrix (BIOM) format or: how I learned to stop worrying and love the ome-ome. Giga Science, 1(1):1\u20136, 2012. R.P. Nugent, M.A. Krohn, and SL Hillier. Reliability of diagnosing bacterial vaginosis is improved by a standardized method of gram stain interpretation. Journal of Clinical Microbiology, 29(2):297\u2013301, 1991. Julia Oh, Sean Conlan, E Polley, Julia A Segre, Heidi H Kong, et al. Shifts in human skin and nares microbiota of healthy children and adults. Genome medicine, 4(10):1\u201311, 2012. J. Oksanen, F.G. Blanchet, R. Kindt, P. Legendre, R. Minchin, R.B. O\u2019Hara, G.L. Simpson, P. Solymos, M.H.H. Stevens, and H. Wagner. vegan: Community Ecology Package, 2012. URL http://CRAN. R-project.org/package=vegan. R package version 2.0-4. F. Pardi and N. Goldman. Resource-aware taxon selection for maximizing phylogenetic diversity. Systematic Biology, 56(3):431\u2013444, 2007. M.N. Price, P.S. Dehal, and A.P. Arkin. FastTree 2\u2013approximately maximum-likelihood trees for large alignments. PLOS ONE, 5(3): e9490, 2010. R Development Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2012. URL http://www.R-project.org/. C.R. Rao. Diversity and dissimilarity coefficients: a unified approach. Theoretical Population Biology, 21(1):24\u201343, 1982. P.D. Schloss, S.L. Westcott, T. Ryabin, J.R. Hall, M. Hartmann, E.B. Hollister, R.A. Lesniewski, B.B. Oakley, D.H. Parks, C.J. Robinson, et al. Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Applied and Environmental Microbiology, 75 (23):7537\u20137541, 2009. C.E. Shannon. A mathematical theory of communication. Bell System Technical Journal, 27(1):379\u2013423, 1948. E.H. Simpson. Measurement of diversity. Nature, 163(4148):688, 1949. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t 15 S. Srinivasan, N.G. Hoffman, M.T. Morgan, F.A. Matsen, T.L. Fiedler, R.W. Hall, F.J. Ross, C.O. McCoy, R. Bumgarner, J.M. Marrazzo, et al. Bacterial communities in women with bacterial vaginosis: high resolution phylogenetic analyses reveal relationships of microbiota to clinical criteria. PLOS ONE, 7(6):e37818, 2012. JM Tanner and RH Whitehouse. Clinical longitudinal standards for height, weight, height velocity, weight velocity, and stages of puberty. Archives of disease in childhood, 51(3):170\u2013179, 1976. M. Vellend, W.K. Cornwell, K. Magnuson-Ford, and A. Mooers. Measuring phylogenetic biodiversity. In Biological Diversity: Frontiers in Measurement and Assessment, A.E. Magurran and B.J. McGill, editors, pages 194\u2013207. Oxford University Press, 2011. Q. Wang, G.M. Garrity, J.M. Tiedje, and J.R. Cole. Na\u0131\u0308ve Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Applied and environmental microbiology, 73(16): 5261\u20135267, 2007. R.M. Warwick and K.R. Clarke. New \u2018biodiversity\u2019 measures reveal a decrease in taxonomic distinctness with increasing stress. Marine Ecology Progress Series, 129(1):301\u2013305, 1995. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Figure 1 \\(g_\\theta\\) curves for various \\(\\theta\\) parameters. as \\(\\theta\\) goes to zero, the \\(g_\\theta\\) converge pointwise to \\(g\\), which is 1 on the : interior of the unit interval and 0 on the boundaries. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Figure 2 Dendrogram relating alpha diversity measures applied to the vaginal dataset. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Figure 3 Comparison of rarefied and unrarefied values of various phylogenetic alpha diversity measures as applied to the vaginal dataset. The value of six alpha measures for each specimen using all available sequences is plotted on the \\(x\\)-axis. The value of the alpha measures for each specimen after a single rarefaction to 523 sequences (the smallest sequence count across specimens) is plotted on the \\(y\\)-axis. The \\(y=x\\) line is shown in blue. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Figure 4 Comparison of diversity between samples from healthy controls, healthy sites of dysbiotic patients, and dysbiotic sites of dysbiotic patients on the oral dataset, using various measures of diversity. \"Shallow\" means a shallow pocket between tooth and gum tissue, while \"deep\" means a sample from a deep pocket between gum tissue that has separated from its tooth. Top row: cluster-based methods. Bottom rows: phylogenetic methods. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Table 1(on next page) Overview of phylogenetic diversity measures used in the text. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t phylogenetic diversity (Faith, 1992) phylogenetic generalization of species count phylo. quadratic entropy (Rao, 1982; Warwick and Clarke, 1995) phylogenetic generalization of the Simpson index phylogenetic entropy (Allen et al., 2009) phylogenetic generalization of the Shannon index qD(T) (Chao et al., 2010) phylogenetic generalization of Hill numbers BWPD1 (Barker, 2002; Vellend et al., 2011) abundance-weighted version of phylogenetic diversity BWPD\u03b8 (this paper) one-parameter family interpolating between PD and BWPD1 TABLE 1. Overview of phylogenetic diversity measures used in the text. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Table 2(on next page) Correlation and predictive performance of the various alpha diversity measures applied to the vaginal data set. Rows are ordered by increasing mean rank across performance measurements. Nugent \\ (R^2\\): \\(R^2\\) value using the measure as a predictor, and the Nugent score as response in a linear model. Amsel accuracy: proportion of specimens with correct BV classification under a leave-one-out cross-validation. Amsel p-value: p-value from a two-sample \\(t\\)-test on values stratified by BV classification. ``OTU'' designates the measure applied to 97\\% clustering groups, and ``Family'' designates taxonomic classification at the family level. Measures described in main text. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Measure Amsel Accuracy Nugent R2 Amsel p-value mean rank PDu 0.834 0.737 1.84E-35 1.3 BWPD0.25 0.836 0.735 1.98E-35 2.0 Simpson (Family) 0.817 0.735 4.11E-33 4.0 BWPD0.5 0.827 0.700 3.33E-33 4.3 Shannon (Family) 0.813 0.724 2.28E-32 5.3 Phylo. entropy 0.831 0.679 1.81E-31 5.7 Chao1 (Family) 0.813 0.704 6.27E-31 7.0 0.5D(T) 0.818 0.658 7.47E-29 8.0 0.25D(T) 0.809 0.682 2.25E-30 8.3 Phylo. quad. entropy 0.813 0.648 7.89E-30 9.0 BWPD1 0.795 0.611 5.38E-28 11.0 Chao1 (OTU) 0.766 0.488 1.64E-23 12.7 ACE (Family) 0.766 0.491 2.82E-11 13.0 ACE (OTU) 0.764 0.469 6.82E-22 13.7 Shannon (OTU) 0.758 0.380 5.27E-16 14.7 Simpson (OTU) 0.697 0.191 1.42E-07 16.0 TABLE 2. Correlation and predictive performance of the various alpha diversity measures applied to the vaginal data set. Rows are ordered by increasing mean rank across performance measurements. Nugent R2: R2 value using the measure as a predictor, and the Nugent score as response in a linear model. Amsel accuracy: proportion of specimens with correct BV classification under a leave-one-out cross-validation. Amsel p-value: p-value from a two-sample t-test on values stratified by BV classification. \u201cOTU\u201d designates the measure applied to 97% clustering groups, and \u201cFamily\u201d designates taxonomic classification at the family level. Measures described in main text. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Table 3(on next page) Predictive accuracy of each measure in the oral dataset and p-value from an ANOVA stratified by disease status and sampling site. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Measure Diseased status accuracy ANOVA p-value mean rank Phylo. entropy 0.791 4.97E-09 1.0 BWPD0.5 0.782 6.50E-09 2.0 BWPD0.25 0.755 7.16E-08 4.0 Phylo. quad. entropy 0.770 2.47E-07 4.0 Simpson (Family) 0.776 1.45E-06 4.0 0.5D(T) 0.734 4.74E-06 6.5 0.25D(T) 0.735 4.33E-05 7.0 PDu 0.691 6.37E-06 8.5 Shannon (Family) 0.734 5.32E-05 8.5 BWPD1 0.698 3.57E-04 9.5 Chao1 (OTU) 0.685 9.94E-04 11.0 ACE (OTU) 0.682 1.30E-03 12.0 Simpson (OTU) 0.676 2.39E-02 13.5 Shannon (OTU) 0.672 1.31E-03 14.0 Chao1 (Family) 0.674 2.64E-01 15.0 ACE (Family) 0.663 1.82E-01 15.5 TABLE 3. Predictive accuracy of each measure in the oral dataset and p-value from an ANOVA stratified by disease status and sampling site PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Table 4(on next page) ANOVA p-values for various diversity statistics applied to the skin microbiome data of oh et al. (2013). : Rows are ordered by increasing mean rank across sites. The same site abbreviations are used as in their paper: Ac, antecubital fossa; N, nares; Pc, popliteal fossa; Vf, volar forearm. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t Ac N Pc Vf mean rank PDu 3.34e-02 4.94e-03 3.03e-03 2.28e-04 4.50 BWPD0.25 4.22e-02 1.32e-03 6.37e-03 5.86e-04 5.25 BWPD0.5 8.54e-02 9.85e-05 3.65e-02 5.85e-03 6.00 Shannon (OTU) 6.61e-02 9.48e-02 9.65e-02 1.05e-05 6.50 Chao1 (OTU) 8.00e-02 6.46e-03 3.98e-03 3.18e-03 6.75 Phylo. quad. entropy 2.52e-01 1.12e-05 4.99e-01 1.67e-01 7.50 Phylo. entropy 1.37e-01 1.15e-03 1.55e-01 2.09e-02 7.75 0.5D(T) 8.91e-01 5.63e-04 3.84e-03 9.09e-01 8.25 0.25D(T) 7.00e-01 2.27e-03 2.35e-03 9.41e-01 8.50 BWPD1 3.09e-01 5.95e-05 6.65e-01 5.41e-01 8.50 0D(T) 4.42e-01 1.05e-02 1.38e-03 9.38e-01 8.75 Simpson (OTU) 9.38e-02 4.01e-01 8.49e-01 1.01e-04 8.75 TABLE 4. ANOVA p-values for various diversity statistics applied to the skin microbiome data of Oh et al. (2012). Rows are ordered by increasing mean rank across sites. The same site abbreviations are used as in their paper: Ac, antecubital fossa; N, nares; Pc, popliteal fossa; Vf, volar forearm. PeerJ reviewing PDF | (v2013:04:390:1:1:NEW 19 Aug 2013) R ev ie w in g M an us cr ip t",
    "url": "https://peerj.com/articles/158/reviews/",
    "review_1": "Lesley Rogers \u00b7 Aug 21, 2013 \u00b7 Academic Editor\nACCEPT\nThank you for making the final corrections.",
    "review_2": "Lesley Rogers \u00b7 Aug 19, 2013 \u00b7 Academic Editor\nMINOR REVISIONS\nThe following points still need to be addressed.\n1. The following sentence in the Introduction is not entirely clear to me. \u201cIn her view, basic level categories may be categorized perceptually before superordinate categories, but with regards to conceptual categories, which are based on kind as opposed to perceptual similarity, more global, abstract categories such as animals, foods etc. may emerge first.\u201d. Please can you improve it?\n2. In the new sentence \u201cWe predicted that orangutans may readily category stimuli from both concrete and intermediate level categories, whereas the gorilla might categorize stimuli more readily at the concrete level.\u201d Correct category to categorize.\n2. You use the word \u2018hoofstock\u2019 in the Introduction but it is not exactly clear what is meant by this.\n3. The first sentence of the Results and Discussion section of Experiment 1 states \u201cAs shown in Figure 2 and confirmed by binomial tests, each subject performed significantly greater than chance..\u201d. Please insert \u2018at a level\u2019 after \u2018performed\u2019.\n4. The second sentence under the heading Experiment 2 doubles up on commas and brackets - ,( and ,). Please choose which of these you want to use.\n4.One of the reviewers said, \u201cI found the tables and figures rather confusing. In table two, the caption says \"percentages\", yet the numbers are all less than 1. And you replied, \u201cA: The numbers are less than one because percentage correct was calculated as a score between 0 and 1 (e.g. 55% = .55).\u201d. Tables 2 and 4 remain a problem because they are not percent scores. Please change to the 0 to 100% range.\n5. In response to a reviewer you changed the following sentence: \u201cAlthough the current study was not designed to address developmental changes in concept acquisition, the fact that at least one gorilla showed greater facility with concrete level categories is interesting and should be explored further.\u201d Replace \u2018should be explored\u2019 by \u2018worth exploring\u2019.",
    "review_3": "Lesley Rogers \u00b7 Aug 15, 2013 \u00b7 Academic Editor\nMINOR REVISIONS\nPlease consider all points raised by the reviewers and address them by making appropriate changes to your manuscript.",
    "review_4": "Lynda Birke \u00b7 Aug 13, 2013\nBasic reporting\nOverall, an interesting paper and generally well written, this study will be of interest to behavioural scientists. I think the paper is publishable, subject to some revisions.\nFirst, at a general level, animals' discriminative skills should be discussed with some reference to perceptual differences. The subjects (great apes) were tested using visual discriminations, based on photos, which is no doubt appropriate for these species. However, the author also makes reference to discriminative skills in other species (eg dogs) whose skills may well have different perceptual bases. The paper should acknowledge that the tests used were specifically testing visual skills, and contextualise discussion accordingly.\nSecond, there should be more discussion of generalisation from the relatively small number of animals used. I am well aware of the difficulties of doing studies with zoo animals - it almost always involves a small N. But here, I felt the paper would benefit from considering how general the demonstrated abilities are (especially for gorillas).\nMore specific points: there should be more detail of how the apes were kept. This is particularly relevant regarding proximity of other species, and the subjects' experience of other primate species. Following on from this, I am sceptical of the implication that specific skills \"pre-existed\" (p. 24) - ie are hard-wired - unless we know much more about these individuals' histories.\nPage 8, paragraph 2 - this was not entirely clear regarding chimps. Please clarify what \"most difficulty with the most abstract distinctions\" means.\nPage 9, lines 4-5 - needs clarification.\nP.12. Please explain why different numbers of trials were used.\np.26. Discussion of how humans classify other species should acknowledge cross-cultural differences in classification.\nI found the tables and figures rather confusing. In table two, the caption says \"percentages\", yet the numbers are all less than 1. The figures do not seem to show any significant improvement over time - please add more information.\nExperimental design\nThe overall design is satisfactory, although I would encourage the author to include more detail of statistics. In particular, the justification of the ANOVA, and the assumptions made in its use, should be clarified (c. page 15)\nValidity of the findings\nSee comments above.\nCite this review as\nBirke L (2013) Peer Review #1 of \"Matching based on biological categories in Orangutans (Pongo abelii) and a Gorilla (Gorilla gorilla gorilla) (v0.1)\". PeerJ https://doi.org/10.7287/peerj.158v0.1/reviews/1",
    "pdf_1": "https://peerj.com/articles/158v0.3/submission",
    "pdf_2": "https://peerj.com/articles/158v0.2/submission",
    "review_5": "Reviewer 2 \u00b7 Aug 6, 2013\nBasic reporting\n(referring to introduction): excellent.\n\nsee general comments also\nExperimental design\nThorough and circumspect and avoiding many potential pitfalls. The researchers took every possible precaution to avoid preferences and ambiguous interpretations of results. Because of the design the experimenter was able to determine that subjects did not rely upon associations they formed during the course of the experiment between particular stimuli pairings and reward in performing this task.\nValidity of the findings\nThe experiments are carefully controlled, the test variables rigorously controlled and the statistical tests appropriate. The results therefore seem robust and the findings very convincing. We are also told the prehistory of experiences of these specific great apes taking part in these experiments reassuring one that these were novel experiences and that it was not prior learning or extensive testing which had produced the current results.\nAdditional comments\nThis is an exciting paper concerned with a possible distinction recognized and known in humans between specific and concrete and global categorizations of objects, object classes and overall classifications without the benefit of language. They used biological classifications of species, family and class. Four orangutans and one gorilla were presented with delayed matching-to-sample tasks and, in two experiments, taken through levels of perceptual matching to abstract matching., i.e. same species group or different families and, finally, fitting individual images into appropriate classes of insect, reptiles, birds or mammals respectively. Impressively, the paper found (for the first experiment) that five of the apes performed above chance levels within the first six sessions. For the second experiment testing the apes\u2019 ability to assign specific photos to classes most apes performed again well above chance, suggesting either that they spontaneously recognised the categories or were able to learn about such categories with relative ease and speed. The authors were cautious and parsimonious in their interpretation and, despite their strong results, resisted the temptation to attribute abstract thinking to the apes but they are saying that biological categorisations do not require specialised taxonomy knowledge but, in this case, a recognition of the basic rules governing the classifications. Their experiments have opened the debate for further studies in this field on how categorisation occurs in pre-linguistic contexts.\nI have very little to add in a critical sense and highly recommend publishing.\nThere are a few minor grammar and other errors in the paper that the author should correct before final submission. One glaring example was the formulation that traits are \u2018shared in common\u2019 \u2013a bad logical error/doubling up: it\u2019s either \u2018have in common\u2019 or \u2018share xyz with someone\u2019\u2026(leaving out \u2018in common\u2019) because when one has certain traits in common with someone else they are, in fact, shared traits (the male stallion problem).\nThere are other small (i.e. readily fixable) issues in presentation: The discussion/conclusion is not as good and tight as the introduction and I would hope that the author could strengthen the discussion by referring back to the introduction in which the categories and distinctions were very well presented indeed and theoretically cogent. Some of the sparkle is gone in the discussion and that is a pity and could be improved simply by using and expanding upon the same sharply defined theoretical concepts and issues as in the beginning of the paper.\nCite this review as\nAnonymous Reviewer (2013) Peer Review #2 of \"Matching based on biological categories in Orangutans (Pongo abelii) and a Gorilla (Gorilla gorilla gorilla) (v0.1)\". PeerJ https://doi.org/10.7287/peerj.158v0.1/reviews/2",
    "pdf_3": "https://peerj.com/articles/158v0.1/submission",
    "review 6": "Reviewer 3 \u00b7 Aug 1, 2013\nBasic reporting\nNo comments\nExperimental design\nNo comments\nValidity of the findings\nNo comments\nAdditional comments\nThis paper investigates the conceptual organization of biological stimuli in perceptual and conceptual categories by non-human primates (gorillas and orangoutangs). In the human literature there has been a debate on the role of language vs. perceptual similarity for the formation of categories, in relation to the level of abstraction implied by a given category. The present study aims at investigating this topic from a comparative perspective, testing non-verbal animals that are phylogenetically close to humans. Categorizing stimuli according to biological taxonomies has hypothesized to be a uniquely human tendency. Contrary to that claim, recent neuroimaging and behavioral data suggested that non-human primates may spontaneously form categories of biological objects in a similar way to what humans do. The current paper investigated the ability to make explicit classifications of natural class distinctions, to determine whether exemplars of more closely related groupings are more readily categorized together compared to more distantly related members of the same class. In the present study, overall 5 subjects were required to match images based on biological classifications at the level of species, family or class. The results of the first experiment indicate that the subjects can form categories at the concrete (species) level, even when confronted with images of unfamiliar primate species. This task however can be solved on the basis of perceptual similarity and does not imply sophisticate abstraction abilities. In Experiment 2 subjects proved able to solve the task also based on a more abstract level of categorization, i.e. using the intermediate categories of taxonomic classes. Moreover, at least orangoutangs proved able to learn intermediate-level categories as readily (or even more readily) as concrete categories, in line with the human developmental literature. On the contrary the gorillas seemed to find it easier to form categories at the concrete level. However, the small sample size prevents from making meaningful comparisons between the performance of the two species (orangoutangs and gorillas). Unfortunately the present study cannot provide any information on the role of experience with different living creatures in the relative facilitation for forming intermediate-level categories displayed by some of the subjects (for example see page 23, first paragraph). This most interesting issue can be assessed only by investigating the performance of subjects reared in strictly controlled environment and/or tested at a very early age after birth. Authors should be careful in their claims on this point (e.g., at page 24). The introduction contains some repeated information, which should be avoided (e.g. the work of Autier-D\u00e9rian et al., 2013 is described two times for no apparent reason). Moreover, the structure of the introduction itself is somehow unclear, some topics seem to assed repeatedly in different parts of the text. It would be helpful to make the introduction more dense and to provide a clearer structure for the reader. It could be also helpful to add, at the end of the introduction, a short paragraph highlighting the main conclusions and open issues originated from the existing literature. Similar problems are present also in the general discussion. The most problematic issue with the present paper is that it is not clear which are the conclusions that can be drawn from the results obtained. It is necessary to highlight the open questions that the study wanted to asses, and how (if) the results obtained provide novel information to answer these questions. Minor comments In the discussion session evidence is reported that categorization of own species images tends to occur at a more concrete level than categorization of distantly related species (page 24 and 25). This is likely to be due to the special status of own-species, which could be perceived as a category per se and elicit more detailed elaboration. It could be also appropriate to discuss this evidence in relation to the \"other species\" effect observed during face perception in humans. The abstract should mention which are the conclusions obtained from the present study. It would be very helpful to include images of the stimuli, since their perceptual appearance is crucial to the interpretation of the results.\nCite this review as\nAnonymous Reviewer (2013) Peer Review #3 of \"Matching based on biological categories in Orangutans (Pongo abelii) and a Gorilla (Gorilla gorilla gorilla) (v0.1)\". PeerJ https://doi.org/10.7287/peerj.158v0.1/reviews/3",
    "all_reviews": "Review 1: Lesley Rogers \u00b7 Aug 21, 2013 \u00b7 Academic Editor\nACCEPT\nThank you for making the final corrections.\nReview 2: Lesley Rogers \u00b7 Aug 19, 2013 \u00b7 Academic Editor\nMINOR REVISIONS\nThe following points still need to be addressed.\n1. The following sentence in the Introduction is not entirely clear to me. \u201cIn her view, basic level categories may be categorized perceptually before superordinate categories, but with regards to conceptual categories, which are based on kind as opposed to perceptual similarity, more global, abstract categories such as animals, foods etc. may emerge first.\u201d. Please can you improve it?\n2. In the new sentence \u201cWe predicted that orangutans may readily category stimuli from both concrete and intermediate level categories, whereas the gorilla might categorize stimuli more readily at the concrete level.\u201d Correct category to categorize.\n2. You use the word \u2018hoofstock\u2019 in the Introduction but it is not exactly clear what is meant by this.\n3. The first sentence of the Results and Discussion section of Experiment 1 states \u201cAs shown in Figure 2 and confirmed by binomial tests, each subject performed significantly greater than chance..\u201d. Please insert \u2018at a level\u2019 after \u2018performed\u2019.\n4. The second sentence under the heading Experiment 2 doubles up on commas and brackets - ,( and ,). Please choose which of these you want to use.\n4.One of the reviewers said, \u201cI found the tables and figures rather confusing. In table two, the caption says \"percentages\", yet the numbers are all less than 1. And you replied, \u201cA: The numbers are less than one because percentage correct was calculated as a score between 0 and 1 (e.g. 55% = .55).\u201d. Tables 2 and 4 remain a problem because they are not percent scores. Please change to the 0 to 100% range.\n5. In response to a reviewer you changed the following sentence: \u201cAlthough the current study was not designed to address developmental changes in concept acquisition, the fact that at least one gorilla showed greater facility with concrete level categories is interesting and should be explored further.\u201d Replace \u2018should be explored\u2019 by \u2018worth exploring\u2019.\nReview 3: Lesley Rogers \u00b7 Aug 15, 2013 \u00b7 Academic Editor\nMINOR REVISIONS\nPlease consider all points raised by the reviewers and address them by making appropriate changes to your manuscript.\nReview 4: Lynda Birke \u00b7 Aug 13, 2013\nBasic reporting\nOverall, an interesting paper and generally well written, this study will be of interest to behavioural scientists. I think the paper is publishable, subject to some revisions.\nFirst, at a general level, animals' discriminative skills should be discussed with some reference to perceptual differences. The subjects (great apes) were tested using visual discriminations, based on photos, which is no doubt appropriate for these species. However, the author also makes reference to discriminative skills in other species (eg dogs) whose skills may well have different perceptual bases. The paper should acknowledge that the tests used were specifically testing visual skills, and contextualise discussion accordingly.\nSecond, there should be more discussion of generalisation from the relatively small number of animals used. I am well aware of the difficulties of doing studies with zoo animals - it almost always involves a small N. But here, I felt the paper would benefit from considering how general the demonstrated abilities are (especially for gorillas).\nMore specific points: there should be more detail of how the apes were kept. This is particularly relevant regarding proximity of other species, and the subjects' experience of other primate species. Following on from this, I am sceptical of the implication that specific skills \"pre-existed\" (p. 24) - ie are hard-wired - unless we know much more about these individuals' histories.\nPage 8, paragraph 2 - this was not entirely clear regarding chimps. Please clarify what \"most difficulty with the most abstract distinctions\" means.\nPage 9, lines 4-5 - needs clarification.\nP.12. Please explain why different numbers of trials were used.\np.26. Discussion of how humans classify other species should acknowledge cross-cultural differences in classification.\nI found the tables and figures rather confusing. In table two, the caption says \"percentages\", yet the numbers are all less than 1. The figures do not seem to show any significant improvement over time - please add more information.\nExperimental design\nThe overall design is satisfactory, although I would encourage the author to include more detail of statistics. In particular, the justification of the ANOVA, and the assumptions made in its use, should be clarified (c. page 15)\nValidity of the findings\nSee comments above.\nCite this review as\nBirke L (2013) Peer Review #1 of \"Matching based on biological categories in Orangutans (Pongo abelii) and a Gorilla (Gorilla gorilla gorilla) (v0.1)\". PeerJ https://doi.org/10.7287/peerj.158v0.1/reviews/1\nReview 5: Reviewer 2 \u00b7 Aug 6, 2013\nBasic reporting\n(referring to introduction): excellent.\n\nsee general comments also\nExperimental design\nThorough and circumspect and avoiding many potential pitfalls. The researchers took every possible precaution to avoid preferences and ambiguous interpretations of results. Because of the design the experimenter was able to determine that subjects did not rely upon associations they formed during the course of the experiment between particular stimuli pairings and reward in performing this task.\nValidity of the findings\nThe experiments are carefully controlled, the test variables rigorously controlled and the statistical tests appropriate. The results therefore seem robust and the findings very convincing. We are also told the prehistory of experiences of these specific great apes taking part in these experiments reassuring one that these were novel experiences and that it was not prior learning or extensive testing which had produced the current results.\nAdditional comments\nThis is an exciting paper concerned with a possible distinction recognized and known in humans between specific and concrete and global categorizations of objects, object classes and overall classifications without the benefit of language. They used biological classifications of species, family and class. Four orangutans and one gorilla were presented with delayed matching-to-sample tasks and, in two experiments, taken through levels of perceptual matching to abstract matching., i.e. same species group or different families and, finally, fitting individual images into appropriate classes of insect, reptiles, birds or mammals respectively. Impressively, the paper found (for the first experiment) that five of the apes performed above chance levels within the first six sessions. For the second experiment testing the apes\u2019 ability to assign specific photos to classes most apes performed again well above chance, suggesting either that they spontaneously recognised the categories or were able to learn about such categories with relative ease and speed. The authors were cautious and parsimonious in their interpretation and, despite their strong results, resisted the temptation to attribute abstract thinking to the apes but they are saying that biological categorisations do not require specialised taxonomy knowledge but, in this case, a recognition of the basic rules governing the classifications. Their experiments have opened the debate for further studies in this field on how categorisation occurs in pre-linguistic contexts.\nI have very little to add in a critical sense and highly recommend publishing.\nThere are a few minor grammar and other errors in the paper that the author should correct before final submission. One glaring example was the formulation that traits are \u2018shared in common\u2019 \u2013a bad logical error/doubling up: it\u2019s either \u2018have in common\u2019 or \u2018share xyz with someone\u2019\u2026(leaving out \u2018in common\u2019) because when one has certain traits in common with someone else they are, in fact, shared traits (the male stallion problem).\nThere are other small (i.e. readily fixable) issues in presentation: The discussion/conclusion is not as good and tight as the introduction and I would hope that the author could strengthen the discussion by referring back to the introduction in which the categories and distinctions were very well presented indeed and theoretically cogent. Some of the sparkle is gone in the discussion and that is a pity and could be improved simply by using and expanding upon the same sharply defined theoretical concepts and issues as in the beginning of the paper.\nCite this review as\nAnonymous Reviewer (2013) Peer Review #2 of \"Matching based on biological categories in Orangutans (Pongo abelii) and a Gorilla (Gorilla gorilla gorilla) (v0.1)\". PeerJ https://doi.org/10.7287/peerj.158v0.1/reviews/2\nReview 6: Reviewer 3 \u00b7 Aug 1, 2013\nBasic reporting\nNo comments\nExperimental design\nNo comments\nValidity of the findings\nNo comments\nAdditional comments\nThis paper investigates the conceptual organization of biological stimuli in perceptual and conceptual categories by non-human primates (gorillas and orangoutangs). In the human literature there has been a debate on the role of language vs. perceptual similarity for the formation of categories, in relation to the level of abstraction implied by a given category. The present study aims at investigating this topic from a comparative perspective, testing non-verbal animals that are phylogenetically close to humans. Categorizing stimuli according to biological taxonomies has hypothesized to be a uniquely human tendency. Contrary to that claim, recent neuroimaging and behavioral data suggested that non-human primates may spontaneously form categories of biological objects in a similar way to what humans do. The current paper investigated the ability to make explicit classifications of natural class distinctions, to determine whether exemplars of more closely related groupings are more readily categorized together compared to more distantly related members of the same class. In the present study, overall 5 subjects were required to match images based on biological classifications at the level of species, family or class. The results of the first experiment indicate that the subjects can form categories at the concrete (species) level, even when confronted with images of unfamiliar primate species. This task however can be solved on the basis of perceptual similarity and does not imply sophisticate abstraction abilities. In Experiment 2 subjects proved able to solve the task also based on a more abstract level of categorization, i.e. using the intermediate categories of taxonomic classes. Moreover, at least orangoutangs proved able to learn intermediate-level categories as readily (or even more readily) as concrete categories, in line with the human developmental literature. On the contrary the gorillas seemed to find it easier to form categories at the concrete level. However, the small sample size prevents from making meaningful comparisons between the performance of the two species (orangoutangs and gorillas). Unfortunately the present study cannot provide any information on the role of experience with different living creatures in the relative facilitation for forming intermediate-level categories displayed by some of the subjects (for example see page 23, first paragraph). This most interesting issue can be assessed only by investigating the performance of subjects reared in strictly controlled environment and/or tested at a very early age after birth. Authors should be careful in their claims on this point (e.g., at page 24). The introduction contains some repeated information, which should be avoided (e.g. the work of Autier-D\u00e9rian et al., 2013 is described two times for no apparent reason). Moreover, the structure of the introduction itself is somehow unclear, some topics seem to assed repeatedly in different parts of the text. It would be helpful to make the introduction more dense and to provide a clearer structure for the reader. It could be also helpful to add, at the end of the introduction, a short paragraph highlighting the main conclusions and open issues originated from the existing literature. Similar problems are present also in the general discussion. The most problematic issue with the present paper is that it is not clear which are the conclusions that can be drawn from the results obtained. It is necessary to highlight the open questions that the study wanted to asses, and how (if) the results obtained provide novel information to answer these questions. Minor comments In the discussion session evidence is reported that categorization of own species images tends to occur at a more concrete level than categorization of distantly related species (page 24 and 25). This is likely to be due to the special status of own-species, which could be perceived as a category per se and elicit more detailed elaboration. It could be also appropriate to discuss this evidence in relation to the \"other species\" effect observed during face perception in humans. The abstract should mention which are the conclusions obtained from the present study. It would be very helpful to include images of the stimuli, since their perceptual appearance is crucial to the interpretation of the results.\nCite this review as\nAnonymous Reviewer (2013) Peer Review #3 of \"Matching based on biological categories in Orangutans (Pongo abelii) and a Gorilla (Gorilla gorilla gorilla) (v0.1)\". PeerJ https://doi.org/10.7287/peerj.158v0.1/reviews/3\n",
    "LLM_extracted_review": "['**Review 1:** Thank you for making the final corrections.\\n\\n**Review 2:**\\n1. The following sentence in the Introduction is not entirely clear to me. \u201cIn her view, basic level categories may be categorized perceptually before superordinate categories, but with regards to conceptual categories, which are based on kind as opposed to perceptual similarity, more global, abstract categories such as animals, foods etc. may emerge first.\u201d. Please can you improve it?\\n2. In the new sentence \u201cWe predicted that orangutans may readily category stimuli from both concrete and intermediate level categories, whereas the gorilla might categorize stimuli more readily at the concrete level.\u201d Correct category to categorize.\\n3. You use the word \u2018hoofstock\u2019 in the Introduction but it is not exactly clear what is meant by this.\\n4. The first sentence of the Results and Discussion section of Experiment 1 states \u201cAs shown in Figure 2 and confirmed by binomial tests, each subject performed significantly greater than chance..\u201d. Please insert \u2018at a level\u2019 after \u2018performed\u2019.\\n5. The second sentence under the heading Experiment 2 doubles up on commas and brackets - ,( and ,). Please choose which of these you want to use.\\n6. One of the reviewers said, \u201cI found the tables and figures rather confusing. In table two, the caption says \"percentages\", yet the numbers are all less than 1. And you replied, \u201cA: The numbers are less than one because percentage correct was calculated as a score between 0 and 1 (e.g. 55% = .55).\u201d. Tables 2 and 4 remain a problem because they are not percent scores. Please change to the 0 to 100% range.\\n7. In response to a reviewer you changed the following sentence: \u201cAlthough the current study was not designed to address developmental changes in concept acquisition, the fact that at least one gorilla showed greater facility with concrete level categories is interesting and should be explored further.\u201d Replace \u2018should be explored\u2019 by \u2018worth exploring\u2019.\\n\\n**Review 3:** Please consider all points raised by the reviewers and address them by making appropriate changes to your manuscript.\\n\\n**Review 4:**\\n1. At a general level, animals\\' discriminative skills should be discussed with some reference to perceptual differences.\\n2. There should be more discussion of generalisation from the relatively small number of animals used.\\n3. There should be more detail of how the apes were kept, particularly regarding proximity of other species and the subjects\\' experience of other primate species.\\n4. Page 8, paragraph 2 - this was not entirely clear regarding chimps. Please clarify what \"most difficulty with the most abstract distinctions\" means.\\n5. Page 9, lines 4-5 - needs clarification.\\n6. P.12. Please explain why different numbers of trials were used.\\n7. P.26. Discussion of how humans classify other species should acknowledge cross-cultural differences in classification.\\n8. I found the tables and figures rather confusing. In table two, the caption says \"percentages\", yet the numbers are all less than 1. The figures do not seem to show any significant improvement over time - please add more information.\\n\\n**Review 5:**\\n1. The discussion/conclusion is not as good and tight as the introduction and could be improved by referring back to the introduction.\\n2. There are a few minor grammar and other errors in the paper that the author should correct before final submission.\\n3. One glaring example was the formulation that traits are \u2018shared in common\u2019 \u2013 a bad logical error/doubling up.\\n4. Some of the sparkle is gone in the discussion and could be improved simply by using and expanding upon the same sharply defined theoretical concepts and issues as in the beginning of the paper.\\n\\n**Review 6:**\\n1. The introduction contains some repeated information, which should be avoided.\\n2. The structure of the introduction itself is somehow unclear; some topics seem to be addressed repeatedly in different parts of the text.\\n3. It would be helpful to make the introduction more dense and to provide a clearer structure for the reader.\\n4. The most problematic issue with the present paper is that it is not clear which conclusions can be drawn from the results obtained.\\n5. The abstract should mention which conclusions were obtained from the present study.\\n6. It would be very helpful to include images of the stimuli, since their perceptual appearance is crucial to the interpretation of the results.']"
}