Darwinius is an adapoid primate from the Eocene of Germany, and its only known specimen represents the most complete fossil primate ever found. Its describers hypothesized a close relationship to Anthropoidea, and using a Saimiri model estimated its age at death. This study reconstructs the ancestral permanent dental eruption sequences for basal Euprimates, Haplorhini, Anthropoidea, and stem and crown Strepsirrhini. The results show that the ancestral sequences for the basal euprimate, haplorhine and stem strepsirrhine are identical, and similar to that of Darwinius. However, Darwinius differs from anthropoids by exhibiting early development of the lower third molars relative to the lower third and fourth premolars. The eruption of the lower second premolar marks the point of interruption of the sequence in Darwinius. The anthropoid Saimiri as a model is therefore problematic because it exhibits a delayed eruption of P2. Here, an alternative strepsirrhine model based on Eulemur and Varecia is presented. Our proposed model shows an older age at death than previously suggested (1.05–1.14 years), while the range for adult weight is entirely below the range proposed previously. This alternative model is more consistent with hypotheses supporting a stronger relationship between adapoids and strepsirrhines.
Adapoids were medium-sized, arboreal euprimates, widespread throughout portions of Europe, Asia, Africa and North America from the Early Eocene to the Late Miocene. Adapoids were a very diverse group, comprising six families and more than 100 species [1–3]. Despite receiving considerable attention in the literature, the evolutionary relationships of adapoids to modern strepsirrhines or haplorhines remain unclear. Currently, two opposing hypotheses predominate: (i) the Adapoid–Anthropoid hypothesis [4–17] and (ii) the Adapoid–Strepsirrhine hypothesis [18–25].
The Adapoid–Strepsirrhine hypothesis has been more broadly accepted for the past two decades, based on the fact that it is recovered in all broadly sampled, recent phylogenetic analyses [22,26–29]. However, the debate surrounding the phylogenetic position of adapoids was thrust back into prominence with the description of the most complete fossil primate skeleton, the caenopithecid adapoid Darwinius masillae. Franzen et al.  proposed a stronger relationship between Darwinius and haplorhines, and Gingerich et al.  later united it specifically with anthropoids. Since then, the biology and the evolutionary relationships of Darwinius and adapoids to modern primates have been discussed extensively in the literature (e.g. [23,24,31–36]). An area that has received much less treatment than the question of Darwinius' phylogenetic position is the age model used  to reconstruct its body mass and age at death. This paper considers the appropriateness of anthropoids generally, and Saimiri specifically, with respect to choosing an appropriate model for Darwninius' development, and also assesses the relevance of growth data to the phylogenetic debates.
1.1 Relevance of dental eruption sequences in primate life history
Life-history analysis assesses the chronology of development and reproduction throughout a species' lifetime, from conception to death. The pattern and timing of tooth emergence known for extant primates can be used to broadly infer the life history of fossil groups [37–39]. This is particularly useful for fossil taxa, in which the dentition is the primary and most reliable gauge of ontogeny . The relevance of the dentition as a source of information to explain life history relies on the fact that it is not strongly influenced by environment . There is also a relationship between the sequence of dental eruption and the overall pace of growth, maturation and other aspects of primate life history . For example, M1 is the first permanent tooth to erupt in primates, and the timing of its emergence is highly correlated with adult brain weight, body weight and probably reflects infant precociality [40,42]. In order to use data from dental eruption sequences in modern taxa to form inferences about a fossil, it is necessary to define the point of interruption of the sequence. Defining this point allows for modern growth trajectory data to be used to determine not only age at death but also other variables such as projected adult weight (e.g. ).
1.2 Dental eruption and life history of Darwinius
The Darwinius specimen is split into two parts: PMO 214.214 and its counterpart WDC-MG-210 .1 The specimens, altogether nicknamed ‘Ida’, come from the site of Messel (near Darmstadt, Germany; 47.5 Ma ). The individual was a weaned and independently feeding juvenile, with an erupted M1 . Based on the absence of a baculum, the specimen has been interpreted as female , although it is possible that this element was lost (e.g. along with the left lower limb below the knee) or had not fully developed at the time of the animal's death.
The preserved dentition (figure 1) allows for several inferences about the permanent dental eruption sequence. We follow Gingerich & Smith's  inference that P2 is an adult tooth, and Franzen et al.  in considering it to be the last tooth to have emerged before death. It is known that variability in the premolar eruption sequence is common in some New World monkeys, such as in Callicebus, Alouatta, Saimiri, Callithrix, Mico, Saguinus and Cebus [45–47]. However, there is no intraspecific variability reported in the presence of P2 in adapoids [48,49], and no reported variation in eruption sequences . Therefore, a priori, there is no empirical basis for believing the eruption of P2 in Darwinius is variable, or that P2 erupted unusually early or late in ‘Ida’. We are additionally assuming that dP2 is replaced by P2 in ‘Ida’, as it is in modern primates that retain this tooth. It is possible that this is an incorrect assumption, and that P2 emerged without having a deciduous precursor, in a manner similar to P1 in many living mammals . However, loss of P1 replacement is inferred to be a very ancient feature in mammalian evolution, based on its absence in even some non-eutherians (e.g. Didelphis ), whereas in all adapoids for which there are adequate data, P2 is known to be replaced [18,44]. As such, arguing for a lack of a deciduous precursor for P2 would require the assumption that this particular lineage developed a peculiar homoplasy, not observed in living primates. Although this is of course possible, in the absence of any data it seems more parsimonious to assume that Darwinius was a typical adapoid and replaced its P2. It is worth noting that in applying a Saimiri model, or indeed a model based on any living primate (e.g. [17,44]), this same assumption is being made.
Following from this inference, the teeth of Darwinius can be divided in two sets: (M1 M2 I1 P2), which are adult and erupted, and (I2 M3 C P4 P3), which are in various stages of development and have yet to erupt, with the point of interruption in the sequence occurring after the eruption of P2 . Using the dental eruption sequence of the New World anthropoid Saimiri as a model, Franzen et al.  estimated an age at death between 9 and 10 months based on the interruption of the dental eruption sequence, and a projected adult weight between 650 and 900 g. For the unerupted teeth, it is also possible to make certain inferences about their likely place in the sequence based on their degree of development. For example, the crown of M3 is completely formed and the tooth is in process of erupting. It was probably only covered by soft tissue, although it still lacks mineralized roots . Therefore, this would place the timing of eruption of M3 early in the sequence.
The goals of the present study are to reassess the age estimate for Darwinius by using ancestral reconstruction of dental eruption sequences, and to further explore the life history of Darwinius. Specifically, we assess the validity of using the anthropoid (Saimiri) model for inferring the age and life history of Darwinius.
2. Material and methods
Ancestral state reconstruction has been embraced by both systematists and evolutionary biologists for many years  as a way of making inferences about the pattern of evolution of particular traits or character complexes. Indeed its implementation can be traced back almost 30 years to the work of Felsenstein . Since then, this method has been applied extensively by many authors in the field of palaeontology (e.g. [54–64]). For the ancestral state reconstruction analysis in this study, dental eruption sequences for 97 fossil and extant taxa were used, including Primates, Scandentia and Soricomorpha (table 1). A matrix of 14 characters related to eruption sequences was created (table 2). The 14 characters used in our matrix constitute the minimum number of informative characters needed to allow the reconstruction of eruption sequences. These characters code for presence or absence of certain teeth, and relative timing of eruption of teeth in different positions. By reconstructing the ancestral states for every character, it is possible, from the information they provide, to infer the order of dental eruption at any particular node on the tree. Although not all of the characters are independent, because of the way they are defined, it is impossible for them to generate conflicting reconstructions. For example, in electronic supplementary material, table S2, the ancestral reconstruction for the euprimate node for character 6 ‘Eruption of I1 relative to the earliest premolar’ recovers both states (i.e. I1 erupting before or after the earliest premolar), instead of one ancestral state. However, because all incisors erupt unequivocally before M2 (character 8, state 3) and all premolars erupt unequivocally after M2 (character 12, state 0), the ancestral state for character 6 resolves as I1 erupting before the earliest premolar.
The cladogram (figure 2) used in this analysis is a supertree based on Marivaux et al. (used for placing Sivaladapis ), Gunnell (for Notharctus ), Arnold et al. (for all living primates [94; v. 3], Silcox et al. (for plesiadapiforms and Tupaia ), Steiper & Seiffert (for Victoriapithecus ), Kay (for Stirtonia and Homunculus ), Kistler et al. (for Hadropithecus and Megaladapis ) and Seiffert et al. (for Adapis, Apidium and Parapithecus ) (see also electronic supplementary material, text S1). Although the trees that we used are considered widely accepted, we acknowledge that there are other phylogenies that might contradict some of the positions in the tree [28,98]. The supertree also reflects discussions on phylogenetic relationships from Strasser & Delson (on Paradolichopithecus and Mesopithecus ), Shoshani et al. (on Archaeolemur ) and Frost (on Kuseracolobus ). Dental eruption sequences exist for two other platyrrhine species that are not included in the analysis, Saguinus nigricollis  and S. graellsi , but Arnold et al.'s  tree did not include them. Because all Saguinus species have very similar dental eruption sequences, we are confident that the exclusion of these two species would not alter the reconstruction of the ancestral anthropoid condition. The outgroups used for the ancestral euprimate node were plesiadapiform taxa. Plesiadapiforms were a group of small, arboreal, archaic mammals widespread during the Palaeocene and the Eocene throughout North America, Europe and Asia. Representatives of plesiadapiforms in our tree include plesiadapids and paromomyids. Here, plesiadapiforms are considered stem primates (following [103–114]). A representative of Scandentia, Tupaia glis, as outgroup for Primates (including plesiadapiformes) has been used. Although ideally we would have also included a dermopteran taxon, no permanent dental eruption sequences for dermopterans have been published. Although the lack of dermopterans in the cladogram would have been problematic if the ancestral primate node was reconstructed, the lowest node reconstructed is the ancestral euprimate node. Because plesiadapiforms are inferred to be closer to Euprimates than any other groups (following Silcox et al. ), under the Outgroup Algorithm  they have the greatest impact on polarizing the euprimate node, so it is very unlikely for dermopterans to produce a new unequivocal resolution at that node. Also, the most comprehensive relevant study supports Sundatheria (the combined clade of Scandentia and Dermoptera) as the living sister group of Primates , rather than Dermoptera alone  or Scandentia alone . This makes Dermoptera no more relevant than the included scandentian in reconstructing the basal euprimate node. The dental eruption sequence of Dymecodon pilirostris (Soricomorpha) has been added as a primitive representation of a mammal dental eruption sequence to further constrain the basal euprimate node. Temporal branch length information for living taxa was taken from Arnold et al. . The branch lengths for fossil taxa originate from many sources and the choices of dates of appearance of lineages are reported and explained in electronic supplementary material, table S4, and electronic supplementary material, figure S2.
Ancestral state reconstructions were executed in the Mesquite v. 3.01 software package , using parsimony. The generalized parsimony algorithm can be applied to optimization of a character of unknown polarity onto a rooted tree, and no additional algorithmic complications are presented by trees containing polytomies . Mesquite also allows missing data when using the parsimony algorithm, but cannot do likelihood calculations with gaps or soft polytomies. Because the data include a significant proportion of relevant fossil taxa, which sometimes produce partially complete eruption sequences, it is important to apply software and an algorithm that can accommodate these limitations, making parsimony the best option. Ancestral permanent dental eruption sequences were reconstructed for five hypothetical ancestors: (i) Euprimates, (ii) stem Strepsirrhini, (iii) crown Strepsirrhini, (iv) Haplorhini, and (v) Anthropoidea. For the ancestral state reconstruction, only lower permanent dental eruption sequences were used, because they are more often reported in the literature, thus increasing the number of taxa available for analysis. Canines were not included in the reconstruction because their time of eruption appears to be influenced by sexual dimorphism [42,82]. Although some primitive primates retain P1, eruption data on first premolars are not included in this analysis because the loss of this tooth early in primate evolution renders this character ambiguous at several nodes. To generate the ancestral eruption sequences, all hypothetical ancestors were assumed to have a P2. The ambiguities in ancestral state reconstructions for the characters ‘Eruption of P2 relative to M3’, ‘Eruption of I1 relative to the earliest premolar’ and ‘Eruption of incisors relative to premolars’ are resolved with restrictions implied by other characters. For the five nodes studied here, the ancestral state reconstructions result in one permanent dental eruption sequence for each hypothetical ancestor. The ancestral reconstruction analysis was also run without including fossil data (see electronic supplementary material, figure S1 and text S2), to evaluate the effect of fossils on the reconstruction.
As discussed in detail below, several lemurids that were similar to Darwinius in body mass and life-history variables were chosen as sources of comparative data. Age-specific body mass data for Eulemur macaco, Eulemur rufus and Varecia variegata (V. v. variegata) were taken from the Duke Lemur Center (DLC) database . Although ‘Ida’ has been inferred to be female , because this inference is based on negative evidence (see above), we have followed the conservative course of including both male and female data in our analysis. None of the living species are known to be sexually dimorphic, and use of just the female data led to extremely similar results (not shown). The percentage of adult body mass achieved at ‘Ida's’ age at death was determined using the age-specific mass estimated from the Lowess regressions from the three lemurid species, and adult mean body mass estimates from each regression. Lowess regressions were obtained using PAST . Dental eruption times for the three lemurids are taken from Smith et al. .
It is important to note that some of the specimens from which the data were collected in Eaglen , and subsequently reported in Smith et al.  are currently assigned to other taxa. What Eaglen  classifies as Eulemur fulvus is currently ascribed to E. rufus according to the DLC, and consequently will be referred to as E. rufus in this paper. Also, the V. variegata sample in Eaglen  is composed of a mixture of individuals of V. variegata and hybrids of V. variegata and V. rubra . Therefore, in this paper we are assigning the dental eruption sequence for these specimens to the genus Varecia in general.
The dental eruption sequence that we use for Saimiri (M1 M2 I1 I2 M3 P4 P2 P3) is the same one as that used by Franzen et al. , as the purpose of this paper is to reassess the viability of the original anthropoid model. However, another eruption sequence is known for Saimiri , which differs in the relative time of eruption of the third molar (M1 M2 I1 I2 P4 P2 P3 M3). While we use Franzen et al.'s  eruption sequence throughout the paper, we discuss in the conclusion how Henderson's  eruption sequence would influence conclusions about the Saimiri model.
3. Results and discussion
3.1 Ancestral reconstruction of permanent dental eruption sequences
Ancestral dental eruption sequences were reconstructed for five nodes (euprimates, stem strepsirrhines, crown strepsirrhines, haplorhines and anthropoids; table 3), based on the ancestral state reconstruction for the 14 characters (see electronic supplementary material, table S2).
The earliest members of Adapoidea and Omomyoidea are very similar in dental morphology , so unsurprisingly there is little variation in eruption sequence inferred for the ancestral nodes. Our ancestral reconstruction suggests two clear trends in the evolution of eruption sequences. The strepsirrhine line is characterized by a primitive dental eruption sequence at the base of stem Strepsirrhini that matches the one inferred for the basal euprimate. Subsequently, this primitive sequence is modified in crown strepsirrhines by the simultaneous eruption of the incisors, along with the canine, in association with the evolution of the toothcomb. The haplorhine line is similarly marked by a primitive basal eruption sequence that resembles the basal euprimate sequence, and then it is characterized by a late eruption of M3 at the base of anthropoids. There are several genera of anthropoids in which M3 erupts comparatively early (Saimiri, Aotus, Pithecia, Pygathrix and Presbytis [65,70,79,83,88,89]), but based on the distribution of this trait on this tree this is inferred here to represent evolutionary events occurring in the context of Anthropoidea, like the loss of M3 in callitrichines. In contrast to the inferred primitive state for anthropoids, Darwinius exhibits early eruption of M3 suggesting that it was more strepsirrhine-like. These results therefore make it difficult to determine the relationship of adapoids to either stem strepsirrhines or basal haplorhines, because they both present the same dental eruption sequence. However, these results are less consistent with the Adapoid–Anthropoid hypothesis because adapoids appear to lack the delay of M3 eruption, a synapomorphic characteristic of primitive anthropoids.
The crown strepsirrhines show another distinctive feature: the presence of a toothcomb. The eruption of a toothcomb results in the almost simultaneous emergence of the incisors and the canine. The eruption of the toothcomb is generally early for crown strepsirrhines [42,68,69,71,73,74], with the exception of Archaeolemur edwardsi . However, the pattern of eruption for the toothcomb does not differ markedly from the eruption pattern of incisors for euprimates, stem strepsirrhines or Darwinius, all of which share an early and contiguous eruption of incisors.
The ancestral reconstruction analysis not including fossil taxa provides similar results (electronic supplementary material, table S3), but there is a substantial difference in the final reconstruction. In the analysis that excludes fossils it is not possible to unequivocally reconstruct the primitive premolar eruption sequence for Anthropoidea, with four different states being inferred to be equally parsimonious. This is particularly problematic because the time of eruption of P2 is crucial for the study of life history of Darwinius specifically. Therefore, the inclusion of fossil data in the analysis is required to resolve relevant ancestral state reconstructions.
3.2 Reassessment of the Saimiri model
Based on the contrasts between anthropoids and Darwinius in the ancestral state reconstruction of dental eruption patterns, Saimiri may not be a good model for the growth of Darwinius, as previously proposed by Franzen et al. . But, contrary to the general anthropoid trend, Saimiri is a fast-growing platyrrhine that, like Darwinius, exibits an early eruption of M3 . The ancestral state reconstruction analysis, however, indicates that this early eruption of the third molar appears secondarily in Saimiri. Because the hypothesized relationships of Darwinius are to stem anthropoids , not Platyrrhini generally, or Saimiri specifically, this similarity would necessarily be a case of homoplasy. Also, cebids generally show a late eruption of P2 [70,79], with the exception of Cebus albifrons , in which it appears to be quite variable . Saimiri especially stands out among the Cebidae for having one of the latest eruptions of the second premolar. Therefore, this pattern contrasts markedly with that observed in Darwinius, in which this tooth is already erupted, with five teeth still remaining unerupted. This has profound implications for calculating the age at death, because it is after the eruption of P2 when the sequence is interrupted by death in Darwinius [17,44]. It is worth noting that Henderson  provides another dental eruption sequence for Saimiri in which the relative time of eruption of the M3 is markedly later. One of the most convincing arguments in favour of the Saimiri model is that squirrel monkeys have, according to the sequence used in Franzen et al. , one of the most strepsirrhine-like dental eruption sequences among anthropoids, precisely because of an earlier relative eruption of M3. In the light of this fact, if Henderson's  sequence is correct, Saimiri would make an even less appropriate model.
By contrast, both stem strepsirrhines and basal haplorhines would make good models for the growth of Darwinius, because of their primitive-looking dental sequences. The only non-anthropoid haplorhine taxa in our sample are tarsiids, which exhibit a dental eruption sequence (table 1) which differs from that inferred for the basal haplorrhine (i.e. extremely early eruption of P2, lack of I2, and simultaneous eruption of M3 and C), making tarsiers a poor choice as model taxa. On the other hand, stem strepsirrhines and Darwinius also share early eruption of M3 and P2. Given that stem strepsirrhines are known only from extinct taxa, without direct information available about their age-specific growth and development, a new growth model based on living strepsirrhines is needed for Darwinius.
Three families of strepsirrhines primitively share a dental eruption sequence similar to that of Darwinius: Lemuridae, Galagidae and Cheirogaleidae [42,68,73,74]. Galagidae and Cheirogaleidae are significantly smaller than caenopithecids . Generally, in mammals, most life-history variables are correlated to body mass , making these very small primates inappropriate models for Darwinius. On the other hand, lemurids exhibit similar body masses to caenopithecids , which makes them a more reasonable model. Dental eruption sequences are known from six lemurids: Lemur catta, Hapalemur griseus, E. rufus, E. macaco, E. mongoz and Varecia sp. . However, L. catta and H. griseus would make poor models for the growth of Darwinius because the eruption of P2 in these two species occurs much later in the sequence, and, as discussed above, this tooth is of critical importance in determining the age at death of this particular specimen. Lemur catta and H. griseus also possess a premolar eruption sequence of 4-3-2, which is derived in the context of Lemuridae, instead of the primitive 2-4-3 pattern found in the rest of lemurids and stem strepsirrhines. Among the three Eulemur species, E. mongoz differs the most from Darwinius in having a late eruption of P2. It would be preferable to apply a model based on species with earlier P2 eruptions. Like E. rufus and E. macaco, Varecia has a 2-4-3 premolar eruption pattern and an early P2 eruption. Additionally, these fast-growing primates are fairly well studied, making them the best living models available for the growth of Darwinius.
The age of eruption of P2 in E. rufus is 1.14 years, 1.05 years in E. macaco and 1.06 years in Varecia. Therefore, the age at death of Darwinius based on this model would have been between 1.05 and 1.14 years, older than previously suggested (9–10 months=0.75–0.83 years ).
For estimating body mass at death, we used ages at death of 1.05 years for E. macaco, 1.14 years for E. rufus and 1.06 years for Varecia. Individuals of E. macaco that aged similarly to Darwinius weigh 75.6% of the adult body mass (approx. 1876 g/2481 g; figure 3a), whereas individuals of E. rufus weigh 78% of the adult body mass (approx. 1699 g/2177 g; figure 3b). Finally, V. variegata achieves 77.7% of the adult body mass at the relevant age (approx. 2763 g/3556 g; figure 3c). Therefore, our lemurid model suggests that ‘Ida’ would have been between the narrow range of 75.6% and 78% of her adult body mass when she died. Franzen et al.  suggested an estimated body weight at death of 485 g. According to the new lemurid model, the projected adult body mass would be between 622 and 642 g. This is consistent with the adult body mass estimations for other caenopithecids (between 500 and 3500 g ), and falls entirely below the estimated adult mass by Franzen et al.  (650 and 900 g).
Our ancestral state reconstruction infers the same dental eruption sequences for basal Euprimates, stem Strepsirrhini and basal Haplorhini. These hypothesized primitive sequences resemble that of Darwinius in the early eruption of M3 and the non-simultaneous eruption of I1-I2-C in contrast to anthropoids and crown strepsirrhines, respectively. The late eruption of M3 in anthropoids and the fact that M3 seems to be the next tooth to erupt in Darwinius at the moment of her death suggests that anthropoids likely do not provide the most appropriate model for estimating growth in adapoids, including Darwinius. The eruption of P2 is important for defining the interruption of the sequence in this particular specimen, and the late eruption of P2 in Saimiri suggests that this genus in particular does not represent a good model for Darwinius.
Our results also suggest that eruption sequences carry useful phylogenetic information. Although variable to some extent, higher level primate taxa (e.g. crown Strepsirrhini, Anthropoidea) can be grouped based on different trends in eruption sequences. Therefore, the study of eruption sequences can contribute to our understanding of primate phylogenetic relationships, in a way that allows for the incorporation of fossil material. In this case, the contrast between the inferred late eruption of M3 in the common ancestor of Anthropoidea, and the advanced stage of development of this tooth in Darwinius, could be interpreted as conflicting with the Adapoid–Anthropoid hypothesis.
The lemurid model for the development of Darwinius proposed in this study does not categorically invalidate the Saimiri model. However, it is an alternative in closer agreement with the more similar dental eruption sequences found in strepsirrhines. Also, it agrees with the currently most widely supported hypothesis for adapoid relationships: the Adapoid–Strepsirrhine hypothesis. This model suggests an older age at death (1.05–1.14 years, depending on the model used) than previously proposed (0.75–0.83 years ). Our model also suggests a narrower range for the projected adult weight (622–642 g) and entirely below the previously proposed (650–900 g ), consistent with caenopithecid range of body masses. Although the current data on lemurid growth are sufficient for certain species of lemurids, better documentation of data on growth, and development, and eruption sequences for more lemurid species would certainly improve the quality of potential new models.
All supporting data are available as the electronic supplementary material.
S.L.T. collected the data and drafted the manuscript. All authors conceived of the study, participated in the design of the study, participated in data analysis, interpreted the data, revised the content of the article and gave final approval for publication.
The authors declare that they have no competing interests.
This work was supported by an NSERC Discovery grant to M.T.S.
We are grateful to S. R. Leigh and J. G. Fleagle for providing access to unpublished data. We thank K. D. Rose, M. A. O'Leary and M. Godinot for their insight on adapoid P2 variability. We thank C. L. Makowski for observations relevant to the assessment of the sex of Darwinius. Thanks to K. Padian and anonymous reviewers for comments that substantially improved this paper.
↵1 PMO: Geological Museum, Natural History Museum, University of Oslo; WDC-MG: Wyoming Dinosaur Center, Messel Grube collection.
- Received July 13, 2015.
- Accepted August 11, 2015.
© 2015 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited.