The preservation of ancient DNA in archaeological fish bone

The field of ancient DNA is taxonomically dominated by studies focusing on mammals. This taxonomic bias limits our understanding of endogenous DNA preservation for vertebrate taxa with different bone physiology, such as teleost fish. In contrast to most mammalian bone, teleost bone is typically brittle, porous, lightweight and is characterized by a lack of bone remodeling during growth. Using high-throughput shotgun sequencing, we here investigate the preservation of DNA in a range of different bone elements from over 200 archaeological Atlantic cod (Gadus morhua) specimens from 38 sites in northern Europe, dating up to 8000 years before present. We observe that the majority of archaeological sites (79%) yield endogenous DNA, with 40% of sites providing samples that contain high levels (> 20%). Library preparation success and levels of endogenous DNA depend mainly on excavation site and pre-extraction laboratory treatment. The use of pre-extraction treatments lowers the rate of library success, although — if successful — the fraction of endogenous DNA can be improved by several orders of magnitude. This trade-off between library preparation success and levels of endogenous DNA allows for alternative extraction strategies depending on the requirements of down-stream analyses and research questions. Finally, we find that — in contrast to mammalian bones — different fish bone elements yield similar levels of endogenous DNA. Our results highlight the overall suitability of archaeological fish bone as a source for ancient DNA and provide novel evidence for a possible role of bone remodeling in the preservation of endogenous DNA across different classes of vertebrates.

In mammals, low bone density is usually associated with poor DNA preservation (Geigl & Grange,82 2018). Archaeological fish bone ( Figure 1A) is typically lightweight, porous, brittle and 83 susceptible to taphonomic damage (Szpak, 2011) and such bone could thus be considered a 84 suboptimal source of aDNA from a mammalian preservation perspective. In contrast to mammals, 85 however, fish bone does not serve as a calcium reservoir under normal conditions (Moss, 1961;86 Witten & Huysseune, 2009). Most higher teleosts, including Atlantic cod (Gadus morhua), lack 87 osteocytes (Kranenbarg et al., 2005;Moss, 1961;Shahar & Dean, 2013;Witten & Villwock, 1997). 88 In acellular fish bone, bone remodeling takes place to a lesser extent and through different cellular 89 and physiological processes (Harland & Van Neer, 2018;Kranenbarg et al., 2005;Witten & 90 Villwock, 1997). An absence of bone remodeling may be important for DNA preservation for 91 several reasons. For example, it has been suggested that an absence of cell lacunae improves the 92 resistance of acellular fish bone to microbial degradation (Szpak, 2011). Moreover, recent 93 evidence indicates that an absence of bone remodeling may aid DNA preservation in specific 94 mammalian bone elements (Kontopoulos et al., 2019). It is therefore possible that the 95 fundamental differences between mammalian and fish skeletal physiology, and especially the lack 96 of bone remodeling in most fish, affects the aDNA preservation potential of archaeological fish 97

bone. 98
Interestingly, multiple studies have reported the successful retrieval of aDNA from archaeological 99 fish bone for a variety of species, locations and age (Oosting et  . Fish aDNA has also been successfully amplified in metagenomic analyses using bulk 108 bone approaches, even from warm tropical climates (Grealy et al., 2016). Finally, high-throughput 109 sequencing (HTS) approaches have yielded high levels (15-50%) of endogenous DNA from a 110 limited number of sites up to one thousand years old (Boessenkool et al., 2017;Star et al., 2017). 111 Despite the clear potential for aDNA preservation in archaeological fish remains, however, no 112 studies have yet investigated the factors that underlie this preservation and it is unclear if the 113 expectation of intra-skeletal variability in DNA preservation observed for mammals is applicable 114 to other vertebrate taxa such as fish. 115 Here, we investigate the preservation of aDNA in archaeological Atlantic cod bones (n = 204) 116 obtained from 38 excavations in northern Europe, dating from 6500 BCE to c.1650 CE (spanning 117 the Mesolithic to early modern periods, Figure 1B, Tables 1 and S1). We use a HTS approach to 118 investigate whether bone element, archaeological site, DNA extraction method, and/or 119 sequencing library preparation protocol can be used to predict library success (i.e., the successful 120 retrieval and amplification of aDNA) and the relative proportion of endogenous DNA. We interpret 121 our results in light of down-stream analytical requirements and provide practical 122 recommendations in order to maximize throughput and inference of whole genome sequencing 123 (WGS) data from ancient fish bone. 124

125
Sample processing and DNA extraction  isopropanol. MinElute purification was carried out using the QIAvac 24 Plus vacuum manifold 138 system (QIAGEN). Parallel non-template controls were included. A subset of 73 samples was 139 subjected to multiple treatments (Table S1). 140 Library preparation, sequencing and read processing were used to fit a Generalized Linear Regression (GLR, endogenous DNA fraction ~ extraction 174 protocol + library protocol + site + bone element). Normality of the data for endogenous DNA 175 content was tested by levels in each of the factors using a Shapiro-Wilk Normality Test. For the 176 GLM and GLR described above several models were run discarding factors that did not show 177 significance in more complex models. Akaike (AIC) and Bayesian Information Criterion (BIC) were 178 used to select the best fitting models. 179

180
A total of 277 sequencing libraries were generated from 204 Atlantic cod bones collected at 38 181 archaeological sites ( Figure 1B, Table S1). Of these, 140 libraries from 29 sites had a minimum 182 concentration of 0.1 ng/μl and were sequenced ( Figure S1, Figure 3A). All libraries showed 183 patterns of DNA fragmentation, fragment length, and deamination rates that were consistent with 184 those of authentic aDNA (Jónsson et al., 2013; Figures S1 and S2, Table S1). Most samples (n = 185 131) were processed once, but a subset of samples (n = 73) was processed using two or more 186 treatment combinations, either using different extraction or library preparation protocols ( Figure  187 3B, Table S1). Bone elements were categorized into three major groups -cranial, postcranial, and 188 pectoral girdle bones ( Figure 2A). The representation of these major groups differs across sites 189 ( Figure 2B, Table S1), which is driven by the availability of elements at the different locations or  Table S1). 203 To statistically infer the most important factors explaining library success we applied two models. 204 First, we focused on the samples that were processed with multiple treatments (n = 73, Figure  205 3B). Second, we incorporated all samples generated from sites with more than three samples (n = 206 191 samples from 27 sites), correcting for multiple treatments by randomly downsampling a 207 single treatment per sample iteratively (i = 100, Figure 3C). The GLM focusing on the samples with 208 multiple treatments (library outcome ~ extraction protocol + library protocol + site + bone 209 element + (1 | Sample)) shows that the outcome is significantly dependent on DNA extraction and 210 library preparation protocols (Table S2) shows a better fit to the data (Table S3). 217 We further assessed whether the same factors affect levels of endogenous DNA for samples (n = 218 124) from 19 locations for which three or more specimens were successfully sequenced by fitting 219 a GLR (endogenous DNA fraction ~ extraction protocol + library protocol + site + bone element). 220 Significantly higher endogenous DNA contents are observed in samples that underwent the DD or 221 BLEDD pre-treatments, compared to a standard DNA extraction ( Figure 4A, Table S4). Given that 222 a number of samples for which DD or BLEDD extraction failed (n = 62) were re-extracted using 223 the standard protocol (figure 3B), such samples may a priori be suspected to have relatively poor 224 DNA preservation. In contrast, library preparation protocol had no significant effect on 225 endogenous DNA content (Table S4). Although postcranial bones tend to have lower levels of 226 endogenous DNA, these differences are not significant, and especially bones from the cranial and 227 pectoral girdle yield comparable levels of endogenous DNA, independent of DNA extraction 228 protocol ( Figure 4B, Table S4). Finally, we observe significant differences in endogenous DNA 229 between sites ( Figure 4C, Table S4) with 8 out of 19 sites yielding samples with high levels (> 230 20%) of endogenous DNA, which includes the oldest excavation (Saevarhelleren, site 7, dated to 231 ca. 6500-6200 BCE). When excluding the non-significant factors from the GLR (bone element and 232 library preparation protocol, Table S5), DNA extraction protocol and site remain significant. The 233 most complex model including all factors shows the best fit to the data (Table S5). 234

235
Here, we present the largest study on DNA preservation in ancient fish bones to date, assessing 236 the effects of bone element, archaeological site, DNA extraction and sequencing library 237 preparation protocols on library success and levels of endogenous DNA. We obtain several 238

conclusions. 239
First, although we did not exhaustively sample all different elements possible, our findings imply 240 that most fish bone elements of sufficient size may be suitable for high-throughput shotgun aDNA 241 analyses. We observe no significant differences in either library preparation success or archaeological fish bone contains more such contaminants than mammalian bone. Fish bone may 293 therefore be less suited for single-tube library preparation protocols than mammalian bone, 294 which can be more efficiently cleaned. 295 Finally, we conclude that a wide range of preservation and excavation conditions can yield high 296 endogenous aDNA preservation in archaeological fish bone. We observe site-specific differences 297 in aDNA preservation, with some sites yielding consistently high rates of library success and levels 298 of endogenous DNA whereas others do not. These site-dependent results make it difficult to 299 predict specific factors underlying sufficient aDNA preservation, as samples from each site are 300 associated with a wide range of different, potentially unknown, pre-and post-excavation 301 taphonomic processes. However, our results confirm that cave sites typically offer ideal 302 conditions for DNA preservation (Bollongino et al., 2008;Hardy et al., 1995), thanks to stable low 303 temperatures and lack of precipitation (Hedges & Millard, 1995). Here, we report the oldest WGS 304 results for archaeological fish bone from the cave site of Saevarhelleren (site 7, Bergsvik et al.,305 2016), which is one of the sites with better DNA preservation despite being up to 8500 years old. 306 In addition to this, we have obtained excellent DNA of bones obtained from dry shell middens (e.g., 307 Orkney Quoygrew, site 12, Harland & Barrett, 2012), as well as bones from waterlogged sediments 308 that were excavated decades ago (e.g., Haithabu Harbour, site 14, Heinrich, 2006).   distinct, individually numbered archaeological sites (see Table 1). Only sites with three or more samples (n) are shown. Note that the distribution of selected bone 368 elements is not necessarily representative of their relative rate of retrieval at specific sites.