New troodontid theropod specimen from Inner Mongolia, China clarifies phylogenetic relationships of later-diverging small-bodied troodontids and paravian body size evolution

Taxon sampling

The dataset used by Pei et al. (2020) consisted of 164 taxa. To explore the interrelationships of the Late Cretaceous troodontids, we expanded the data matrix to 171 taxa by incorporating Almas (Pei et al., 2017b), MPC-D100/1126+D100-3500 (Pei, 2015), Gobivenator (Tsuihiji et al., 2014), Talos (Zanno et al., 2011), Linhevenator (Xu et al., 2011a), Philovenator (Xu et al., 2012), and LH PV39 (the present study). Characteristic scores of these taxa except LH PV39 were copied from (Pei, 2015), and scores of the newly added characteristics 854 and 855 were coded by referring to the original descriptions of these taxa (see Section Character sampling and scoring for more information). Other troodontids, including Urbacodon (Averianov and Sues, 2007) and Latenivenatrix (van der Reest and Currie, 2017), are extremely fragmentary and are, therefore, not included in this analysis. In addition, we excluded Jinfengopteryx from our analysis, which was scored based on inadequate anatomical descriptions (Ji, 2005; Ji and Ji, 2007) and was the most unstable OTU in terms of its position when the above seven new taxa were included.

Character sampling and scoring

Note that character 631 (relative size of the posterior surangular foramen) in Pei (2015) was not included in Brusatte et al. (2014) or in Pei et al. (2020); similarly, character 853 (morphology of palatine pneumaticity) of Brusatte et al. (2014) was removed from the matrix of Pei (2015). No reason was given for these adjustments (Pei, 2015). This led to the scores of Almas, MPC-D100/1126+D100-3500, Gobivenator, Talos, Linhevenator, and Philovenator to be shifted backward by one position, as a whole, from character 632 to 853 when copying them from Pei (2015) to the present matrix. To be consistent with the dataset of Pei et al. (2020), we adjusted our matrix by removing the score of character 631 from Pei (2015) and adding back the score of character 853 from Brusatte et al. (2014) for Almas, MPC-D100/1126+D100-3500, Gobivenator, Talos, Linhevenator, and Philovenator.

In addition, a total of nine modifications were made based on the dataset used by Pei et al. (2020) obtained primarily from recent descriptive studies of relevant taxa. Our modifications include the following: reinterpreting a large maxillary fenestra occupying more than half of the area anterior to the internal antorbital fenestra that is in Xixiasaurus, Zanabazar, Saurornithoides and Byronosaurus (Character 27: 2) (Makovicky et al., 2003; Bever and Norell, 2009; Norell et al., 2009; Lü et al., 2010); reinterpreting the external naris in Byronosaurus that extends posteriorly as far as to the anterior margin of the antorbital fossa (Character 23: 1) (Makovicky et al., 2003; Bever and Norell, 2009); interpreting a distinct ventral depression located between the parapophyses of cervical vertebrae in Sinornithoides (Character 787: 1) (Currie and Dong, 2001); estimating the sacral number to 6 in Saurornithoides (Norell et al., 2009) and Buitreraptor (Character 108: 1) (Gianechini et al., 2018; Novas et al., 2018); and identifying the unfused sacral neural spines in Buitreraptor (Character 670: 0) (Gianechini et al., 2018). These modifications are listed in the Supporting Information.

In addition to the 853 phenotypic characters employed by Pei et al. (2020), the current dataset includes two newly added characters, both describing the morphology of the sacral vertebrae:

854: sacral centra: spool-like (0), or dorsoventrally compressed and plate-like (1), or transversely compressed (2). (newly added).

855: sacral centra: similar in width (0), or the maximum width occurs at the mid-length of the centra (1), or the maximum width occurs at both ends of the centra (2), or centra width constantly decreases posteriorly (3), or centra width constantly increases posteriorly (4). (newly added).

To explore all the possible phylogenetic positions of LH PV39, we conducted five parsimony-based phylogenetic analyses by scoring LH PV39 using different strategies. First, we included LH PV39 as a separate operational taxonomic unit (OTU) and scored all available characters, with the aim of exploring the phylogenetic position of LH PV39 and its potential influence on the interrelationships of troodontids (Fig. 6a). Next, we explored the contributions of the sacral and cervical characters to the placement of LH PV39 by rescoring sacral (Characters 108, 109, 110, 111, 112, 316, 670, 672, 673, 674, 854, and 855) and cervical (Characters 90, 93, 94, 95, 96, 97, 98, 99, 104, 260, 460, 652, 660, 661, 662, 663, 664, 665, 786, 787, 788, and 832), respectively, of LH PV39 to unknown (“?”) (Fig. 6b,c). Because LH PV39 was recovered from a locality that is relatively close geographically to that of Philovenator, and size comparisons of Philovenator and LH PV39 do not preclude the referral of LH PV39 to the latter taxon (Fig. 2b), we also combined the scores of Philovenator and LH PV39 into a composite OTU named “Philovenator+LH PV39” to test whether the referral of LH PV39 to Philovenator alters the placement of the latter taxon (Fig. 6d). Moreover, to test whether the paravian interrelationships were significantly altered when LH PV39 is referred to as Linhevenator (Fig. 6e), we combined the scores of Linhevenator and LH PV39 into a composite OTU named “Linhevenator+LH PV39”, despite the considerably larger inferred size of the former taxon relative to LH PV39 (Fig. 2).

Cervicals (Figs 3 and 4)

The exact positions of the five relatively complete cervical vertebrae of LH PV39 are difficult to determine given that relatively complete and well-described vertebral series are rarely known for troodontids (Gao et al., 2012; Shen et al., 2017a; Shen et al., 2017b; Xu et al., 2017). Assuming 10 cervicals were present as is the case in most other coelurosaurs (Currie and Chen, 2001; Hwang, 2004; Dal Sasso and Maganuco, 2011; Pei et al., 2014; Pei et al., 2017a; Gianechini et al., 2018), the five relatively complete cervical vertebrae of LH PV39 are tentatively identified as C4, C5, C6, C8, and C9, respectively (Figs 3 and 4). These identifications are based on the increasingly greater separation of diapophyses and parapophyses, and on the progressive reduction of offset between the anterior and posterior articular surfaces of the centrum posteriorly across the column. A partial left prezygapophysis in articulation with the diapophysis probably originates from C7 (personal observation). This element is briefly described and will not be discussed further.

All cervical vertebrae have a neural arch that is completely fused to the centrum without bearing any trace of an intervening suture (Figs 3 and 4). The centra of C4-C6 are approximately 2 to 3 times longer than deep, whereas this value is less than 2 in C8 and C9. This suggests that the anterior cervicals are more elongated than the posterior ones as in other paravians (Currie and Dong, 2001; Pei et al., 2017a; Shen et al., 2017b; Xu et al., 2017). In most paravians, the middle cervicals (C5-C7) tend to be the most elongated elements in the neck (Gianechini et al., 2018). As cervicals are shorter both anterior and posterior to this point (Table 1), the peak centrum length of LH PV39 is likely to be present at C5, as in other troodontids (Shen et al., 2017b; Xu et al., 2017). It should be noted that the centra of C4 through C6 extend posteriorly beyond the postzygapophyses of the corresponding vertebra as in Troodon (Makovicky and Norell, 2004) and Sinornithoides (Currie and Dong, 2001), whereas they are slightly shorter than the postzygapophyses in more posterior cervicals (e.g., C8 and C9). As in Saurornithoides (Norell and Hwang, 2004) and Mei (Gao et al., 2012), the moderately concave anterior articular surfaces of the centra are wider than they are deep, in contrast to the relatively flat posterior surfaces (Fig. 3). Therefore, the centra are amphiplatyan, given that the anterior surfaces are not as concave as those of the typical procoelous condition. In C4-C6 (Figs 3 and 4), the anterior surface of the centrum is dorsally offset relative to the posterior surface of the corresponding centrum as in other non-paravian dinosaurs (Brochu, 2003; Brusatte et al., 2012). This offset is progressively weaker in more posterior cervical vertebrae, suggesting that the neck is naturally assumed in a sigmoid position. In addition, when the centrum is horizontally oriented, the anterior and posterior surfaces of the centra of C4-C6 face anteroventrally and posterodorsally, respectively. This is consistent with the interpretation of a curved neck, but differs from the condition in Buitreraptor in which the anterior and posterior surfaces of the middle cervical centrum face anteroventrally and posteriorly, respectively (Gianechini et al., 2018).

In all cervicals, the parapophyses are located at the anteroventral corner on the lateral aspect of the centrum (Figs 3 and 4), unlike in Mei, where the parapophyses are located as high as the dorsoventral midpoint of the centrum (Gao et al., 2012). The morphology and orientation of the parapophyses change along the neck, as is typical in dinosaurs (Madsen, 1976; Brochu, 2003; Brusatte et al., 2012). In C4-C8, the parapophyses are triangular, plate-like structures in lateral view. However, they project lateroventrally in C4 (Fig. 3a) in anterior view, whereas they extend primarily laterally with their terminal ends located only slightly lower than the centrum in C8 (Fig. 4b). In C9, the parapophyses project entirely laterally, and they are significantly reduced in size compared to those of the anterior cervicals (Fig. 4c). The parapophyseal articular surfaces for the capitulum are strap-like, and become progressively more anteroposteriorly elongate posteriorly from C4 through C8. In contrast, they are ovoid and concave in C9 (Fig. 4c). The left parapophysis of C5 is excavated by a deep fossa (Fig. 3b). Such a foramen is absent from the opposite side; therefore, we interpret this feature as a postmortem artifact or trace of scavenging by osteophagous insects.

The lateral surface of each cervical centrum is moderately concave and excavated by a pneumatic foramen (sensu pleurocoel of some previous authors (Currie and Dong, 2001; Gao et al., 2012; Gianechini et al., 2018; Hartman et al., 2019)) that is located immediately above the base of the parapophysis. This foramen is also present in other troodontids, including Saurornithoides (Norell and Hwang, 2004), Mei (Gao et al., 2012), Liaoningvenator (Shen et al., 2017b), Hesperornithoides (Hartman et al., 2019), and the unenlagiid Buitreraptor (Gianechini et al., 2018). However, two pneumatic foramina are present on each side of at least the last cervical centrum of Buitreraptor (Gianechini et al., 2018). A similar configuration is also present in Sinornithoides (Currie and Dong, 2001), with the anterior foramen located more posteriorly than in LH PV39. The pneumatic foramen is slitlike in C4 through C6 (Figs 3 and 4), without presenting a shallow fossa surrounding it. However, it becomes rounded and more deeply excavated into the centrum posteriorly across the cervical series. An additional shallow depression is present posterior to the pneumatic foramen and close to the neurocentral junction (Fig. 3a). This shallow depression is more discernible in C4 and C5 than in the more posterior cervicals. Furthermore, it is slightly deeper anteriorly, and it becomes dorsoventrally expanded posteriorly, giving it a triangular profile (Fig. 3). This depression is even more elongated anteroposteriorly than the pneumatic foramen in C4 and C5 (Fig. 3), whereas it becomes more oval, shallower, and less distinct than the pneumatic foramen in C6, C8, and C9 (Fig. 4).

The ventral surfaces of the cervical centra are generally smooth, and there is no trace of a keel or hypapophysis, which is in contrast to the condition in Troodon (Makovicky and Norell, 2004), Saurornithoides (Norell and Hwang, 2004), Mei (Gao et al., 2012), Sinornithoides (Russell and Dong, 1993), and the unenlagiid Buitreraptor (Gianechini et al., 2018). This surface is anteriorly bordered on each side by a ridge that extends from the parapophysis and becomes confluent posteriorly with the centrum (Fig. 3a). Because the parapophyses are lateroventrally projected in anterior cervicals, the ventral margins of C4 and C5 appear strongly concave in lateral view even though the ventral surface itself is flat. However, the ventral surface is concave in more posterior cervicals, namely C6 through C9 (Fig. 4). The anterior portion of the ventral surface between the parapophyses bears a shallow triangular fossa (Fig. 4a,b), with one of the triangular vertices towards posteriorly. In C6 through C9, these fossae are laterally bounded by a moundlike carotid process on each side (Fig. 4), which has been recovered as a paravian synapomorphy (Russell and Dong, 1993; Makovicky, 1995; Currie and Dong, 2001; Gao et al., 2012; Gianechini et al., 2018; Pei et al., 2020; Pittman et al., 2020). Additionally, the carotid processes are more prominent in C9 than those present in C6 and C8 (Fig. 4c), and the distance between the paired carotid processes increases posteriorly along the neck, unlike in Sinornithoides in which the distance between the paired carotid processes decreases posteriorly across the cervical series (Currie and Dong, 2001).

Although all cervical centra of LH PV39 are fused to their respective neural arches, relatively complete neural arches are only preserved in C4 and C5 (Fig. 3a,b). In dorsal view, the neural arch is transversely narrow at its midpoint, and the dorsal surface lateral to the neural spine is moderately concave (Fig. 3a,b). The neural spines of C4, C5, and C8 are preserved, but none of them is complete. Judging from the preserved bases, the neural spines are relatively thin and become shorter in more posterior cervicals. A pair of spinoprezygapophyseal laminae extends anterolaterally from the base of the neural spine, and becomes confluent with the medial margin of the prezygapophyses (Fig. 3b). Proximally, the spinoprezygapophyseal laminae demarcate the dorsolateral margins of the spinoprezygapophyseal fossa (sensu prespinal fossa; (Brusatte et al., 2012)), located directly anterior to the neural spine. The spinoprezygapophyseal fossa is ventrally separated from the neural canal by the medially convergent intraprezygapophyseal laminae, which protrude slightly ventrally along the midline due mainly to the postmortem deformation. These elements give the neural canal a heart-shaped appearance (e.g., C5; Fig. 3b). The neural canal is wider than it is deep, and it becomes progressively larger posteriorly along the neck.

Portions of the pre- and postzygapophyses are present in all cervicals. Although most of the left prezygapophysis was eroded in C4 and C5, the right prezygapophyses are particularly well-preserved (Fig. 3a,b). In dorsal view, the prezygapophyses are slightly more widely spaced relative to the central midline than the postzygapophyses (Fig. 3a,b). The prezygapophyses extend primarily anterolaterally, and those in C5 are considerably more slender and anteriorly extended relative to those in C4 (Fig. 3). The prezygapophyseal articular surface is ovoid and smooth, with the long axis trending primarily anteroposteriorly (Fig. 3a,b). In lateral view, the articular surface is oriented anterodorsally, facing more anteriorly in C4 than in C5 (Fig. 3a,b). This suggests that the neck would have assumed a natural curve at this position.

In anterior view, the centroprezygapophyseal lamina extends medioventrally from the medial aspect of the prezygapophysis and finally contributes to the anterolateral margin of the neural canal (Fig. 3b). In addition to this lamina, the prezygodiapophyseal lamina extends posteroventrally from the ventral aspect of the prezygapophysis and connects the diapophysis (Fig. 3a). This strut is transversely thick where it coalesces with the prezygapophyseis. Further, it becomes thinner as it extends posteroventrally, forming the anterolateral margin of the large triangular flange of the diapophysis and also contributing to the lateral margin of the anteriorly-facing prezygapophyseal centrodiapophyseal fossa (Wilson, 1999; Brusatte et al., 2012), lateral to the neural canal. This pocket is most prominent in C4 through C6 (Figs 3a and 4a) but it remains unclear whether the pocket is also present in posterior cervicals.

The articular surfaces of the diapophyses are strap-like, with the long axis trending primarily anteroposteriorly in anterior cervicals before becoming ovoid in C7. Moreover, this articular surface faces lateroventrally in C4 through C6, and it closely approaches the parapophysis in anterior cervicals (Figs 3a,b and 4a). In C9, this articular surface is located posterior to the parapophysis at the approximate height of the neurocentral suture and faces ventrally (Fig. 4c). This suggests that the length, position, and orientation of the diapophysis change posteriorly across the cervical series in LH PV39, as they do in other dinosaurs (Brochu, 2003; Brusatte et al., 2012).

In addition to the prezygodiapophyseal lamina, three more discrete laminae extend from the diapophysis in C4 through C6. Two of these connect with the centrum and one with the postzygapophysis. The anterior centrodiapophyseal lamina starts from the anteromedial corner of the diapophysis and extends medially until it reaches the neurocentral junction (Fig. 3a). This lamina demarcates the ventral margin of the prezygapophyseal centrodiapophyseal fossa, which is discernible in anterior view of C4 through C6 but is obscured by the pendant diapophysis when viewed laterally (Fig. 3a–c). The posterior centrodiapophyseal lamina extends posteriorly from the posteromedial corner of the diapophysis, overhangs the pneumatic foramen above the parapophysis, and gently coalesces with the centrum at approximately the anteroposterior midpoint of the centrum (Fig. 3a). Dorsoventrally broad, at its origin, this lamina becomes a thin osseous strut posteriorly. The postzygodiapophyseal lamina links the diapophysis with the postzygapophysis. This lamina, along with the prezygodiapophyseal lamina, defines the lateral margins of the large triangular diapophysis in all cervicals. There is a deeply excavated pneumatic foramen posterior to the diapophysis and on the lateral aspect of the neural arch that is dorsally and ventrally demarcated by the postzygodiapophyseal and the posterior centrodiapophyseal laminae, respectively (Fig. 3a,b). This feature is also present in most other troodontids (Norell and Hwang, 2004; Shen et al., 2017b), but it seems to be absent in Mei (Gao et al., 2012). CT images show that this foramen does not deeply excavate the neural arch in C4 (personal observation), rather, it invades the posterior aspect of the diapophysis in C5 and C6.

Complete paired postzygapophyses are known only in C5 (Fig. 3b), and the right postzygapophysis of C4 is partially preserved (Fig. 3a). In dorsal view, the postzygapophyses do not extend posteriorly to the extent of the corresponding centrum in C4 and C5, whereas they do in C8 (Fig. 4b). In C4, the postzygapophyseal articular surface is ovoid and moderately concave (Fig. 3a). This is in contrast to the condition in C5, in which it is relatively flat and transversely narrow (Fig. 3b). In addition, this surface faces primarily posteroventrally and only slightly laterally in C4, whereas in C5, it faces posteroventrally to the same degree that it faces laterally. Although the postzygapophyses have been completely eroded in C9, on the left side of the posterior margin of the neural canal, there is a smooth surface that is transversely wide and extends to the lateral surface of the neural arch. However, since a similar structure is not present in the opposite side of this vertebra nor on other cervical vertebrae of LH PV39, this is likely to be a pathological structure or merely a preservational artifact.

The postzygapophyses do not connect with the main body of the neural arch through a bony neck (Fig. 3a,b), which differs from the prezygapophyses. In dorsal view, the postzygodiapophyseal and intrapostzygapophyseal laminae contribute to the lateral and posterior margins of the postzygapophyses, respectively (Fig. 3a,b). In posterior view, the intrapostzygapophyseal lamina from both sides converge along the midline, where they protrude slightly ventrally. This protrusion is due primarily to postmortem deformation, and it gives the neural canal a heart-shaped appearance (Fig. 3b). Here, the intrapostzygapophyseal laminae also form the bottom of a well-developed spinopostzygapophyseal fossa (or postspinal fossa (Wilson et al., 2011)), marking the insertion point of the interspinous ligament (Fig. 3). The spinopostzygapophyseal lamina is raised dorsally as a slight lip that contributes to the medial margin of the postzygapophysis, and it connects the neural spine and the postzygapophysis (Fig. 3b). In C4, the spinopostzygapophyseal fossa is more prominent and elongated anteroposteriorly than the spinoprezygapophyseal fossa. However, in C5, the spinoprezygapophyseal fossa is deeply excavated into the neural spine and approximately as large as the spinopostzygapophyseal fossa. Further comparison is precluded by incomplete preservation of spinopre- and spinopostzygapophyseal fossae in C8.

As in Mei (Gao et al., 2012) and Saurornithoides (Norell and Hwang, 2004), an epipophysis is present on the dorsal surface of the postzygapophysis perhaps indicating relatively strong neck musculature in these taxa. In dorsal view, epipophyses in LH PV39 are located at the center of the postzygapophysis in C5 (Fig. 3b), whereas it is located much closer to the posterolateral margin of the postzygapophyses in C4 (Fig. 3a). This suggests that the position of the epipophysis varies across the neck. A weak ridge extends anteriorly from the epipophysis in C4 for a distance that is approximately the same length as the epipophysis itself, but it does not form a lamina that connects anteriorly to the prezygapophysis as in Alioramus (Brusatte et al., 2012).

The available cervical series of LH PV39 also includes a partial left prezygapophysis in association with diapophysis (personal observation). Because the prezygapophysis and diapophysis are closely situated, this prezygapophysis likely derives from a cervical posterior to C6. The articular surface of this prezygapophysis is ovoid, flat, and slightly more expanded than that in C9, and its base is also longer than that of C9 anteroposteriorly. Therefore, given that the bases of the prezygapophyses are preserved in both C8 and C9, this prezygapophysis most likely originates from C7.

Synsacrum (Fig. 5a–g)

The synsacrum of LH PV39 consists of five vertebrae that are fully co-ossified, although residual intercentral contacts are still visible in some areas (Fig. 5a–c). Among troodontids, Late Cretaceous taxa such as Saurornithoides (Norell et al., 2009), Troodon (Norell et al., 2009), Gobivenator (Tsuihiji et al., 2015), Latenivenatrix (van der Reest and Currie, 2017), Zanabazar (Norell et al., 2009) and an unnamed specimen collected from Two Medicine Formation (van der Reest and Currie, 2017) each has a sacrum consisting of six vertebrae. In contrast, there are only five sacral vertebrae in the Early Cretaceous troodontid Mei (Gao et al., 2012), Sinusonasus (Xu and Wang, 2004), Daliansaurus (Shen et al., 2017a), and Sinovenator (Xu et al., 2002) and in many other paravians including Archaeopteryx, Anchiornis, Velociraptor, etc. (Ostrom, 1976; Norell and Makovicky, 1997; Wellnhofer, 2009; Xu et al., 2011b; Turner et al., 2012; Godefroit et al., 2013; Gianechini et al., 2018). A sacrum consisting of five vertebrae is previously known in only one Late Cretaceous troodontid, specifically the unnamed MPC-D100/1126 (Pei, 2015), suggesting that the sacral vertebral count is highly diverse among Late Cretaceous troodontids. Character optimization indicates that five sacral vertebrae are the plesiomorphic condition for troodontids, and the addition of a sixth sacral vertebra occurred one or more times during troodontid evolution. Gianechini et al. (2018) suggested that the number of sacral vertebrae could vary among the specimens of the unenlagiid Buitreraptor, and it is likely to be an ontogenetically variable feature. However, the number of vertebrae is determined at the early stages of embryonic development, and is unlikely to change during the late stages (Mallo et al., 2010). Therefore, if they were correctly identified, the different sacral vertebral counts among specimens of Buitreraptor are likely due to intraspecific variation.

All sacral centra are well-preserved. In ventral view, each sacral centrum is constricted at its midpoint (Fig. 5a–c). CT images show that they are extremely compressed dorsoventrally (Fig. 5a–e) and not spool-like elements, unlike most other theropods (Brochu, 2003; Norell et al., 2009; Zanno et al., 2011; Brusatte et al., 2012; Sues and Averianov, 2016; van der Reest and Currie, 2017). Dorsoventrally compressed sacral centra are also present in the troodontid Sinovenator (Xu et al., 2002), the dromaeosaurids Rahonavis (Forster et al., 1998), Buitreraptor (Gianechini et al., 2018; Novas et al., 2018), Mahakala (Turner et al., 2007b; Turner et al., 2011), and in the oviraptorosaurs Microvenator (Makovicky and Sues, 1998), Avimimus (Funston et al., 2019), Chirostenotes (Currie and Russell, 1988; Sues, 1997), and the embryonic oviraptorids (Wang et al., 2016). In LH PV39, a broken surface on the left side reveals that the centrum of S2 is a plate-like element with a dorsal aspect that is excavated by shallow depressions, suggesting that this region has been substantially eroded. CT images reveal that the centra of the non-eroded S4 and S5 are also dorsoventrally compressed, demonstrating that this likely represents an actual feature of LH PV39 and is not a result of postmortem deformation or erosion.

The anteroposterior length of the individual sacral centra decreases posteriorly from S1 through S3, reaching a minimum length of 7.25 mm in S3 before increasing again posteriorly (Table 1). Variation in length among sacral vertebrae is extensive among paravians; in most paravians, the sacral vertebrae are subequal in length (e.g., Anchiornis (Pei et al., 2017a), Mei (Xu and Norell, 2004), Sinovenator (Xu et al., 2002), and Zanabazar (Norell et al., 2009)). The six sacral vertebrae of Mahakala are similar in length, except for the first and last vertebrae, which are significantly longer than those in between (Turner et al., 2011). Therefore, LH PV39 resembles Mahakala in both the compressed morphology and the length variation of the sacral centra.

Although the vertebrae have been completely fused, the residual intercentral contacts remain visible as the contact regions expand beyond the centers of the centra (Fig. 5c). The transverse widths of the centra increase posteriorly in S1 through S3 (Table 1), reaching a maximum width of 16.29 mm in S3, and then decrease posteriorly as in Sinovenator (Xu et al., 2002), Microraptor (Xu et al., 2000), and Archiornis (Pei et al., 2017a). Despite the fact that the intercentral junctions of S1 and S2, and S4 and S5 are of similar width, the anterior articular surface of the S1 centrum is transversely narrower than the posterior surface of the corresponding centrum. In contrast, the anterior surface of S4 is transversely wider than the posterior surface of the respective centrum (Fig. 5c). This suggests that the sacral central width increases and decreases dramatically in S1 and S4, respectively. Among known sacrals, the last centrum (S5) is the narrowest (Fig. 5c).

As in other theropod dinosaurs, the ventral surface of the centrum in S1 is smooth and gently convex, with no sign of keels or grooves. However, a shallow groove extends posteriorly from the junction of S1 and S2 until the midlength of S5, cutting through the expanded intercentral junctions between S2 and S3, S3 and S4, and S4 and S5 (Fig. 5c). The transverse width of this groove varies little over its length, which is approximately one-third of the last sacral's width (S5). A similar midline groove is also present in Latenivenatrix (van der Reest and Currie, 2017), Mahakala (Turner et al., 2011), Balaur (Brusatte et al., 2013), Velociraptor (Brusatte et al., 2013), Buitreraptor (Gianechini et al., 2018), and Bambiraptor (Burnham, 2004), but is absent in Saurornitholestes (Norell et al., 2009; van der Reest and Currie, 2017), Archiornis (Pei et al., 2017a), and the only known sacral vertebra of Talos (Zanno et al., 2011). In Zanabazar, the ventral midline groove can be discerned only near the intercentral junctions (Norell and Hwang, 2004). The extent of this groove also varies among paravians. It marks only the second through the fourth sacrals in Balaur and Velociraptor (Brusatte et al., 2013), whereas it starts from the posterior half of the second sacral in Mahakala (Turner et al., 2011). In Buitreraptor, this groove extends posteriorly until the midlength of the last sacral (Gianechini et al., 2018). In Latenivenatrix (van der Reest and Currie, 2017), this groove is present in S3 through S5, whereas it is present only on the third sacral in Bambiraptor (Burnham, 2004). These regions are fairly smooth in LH PV39, unlike in other theropod dinosaurs for which the ventral surfaces of the centra are marked by a series of anteroposteriorly trending striations adjacent to the intercental contacts (Brusatte et al., 2012).

Articular surfaces of the sacral centra cannot be discerned, and even CT images could not reveal details of the intercentral junctions. This confirmed that the sacrals have been completely fused. Therefore, only the anterior and posterior articular surfaces of S1 and S5, respectively, are exposed (Figs. 5d,e). The anterior surface of S1 is ovoid, nearly flat, and approximately twice as wide transversely as deep, similar to the condition in Buitreraptor (Gianechini et al., 2018; Novas et al., 2018) (Fig. 5d). In lateral view, this surface is almost straight vertically, and is slightly offset ventrally relative to the intercentral junction of S1 and S2 (Fig. 5b). In contrast, the trapezoidal posterior surface of S5 is concave, with a straight ventral margin that is transversely wider than the dorsal margin (Fig. 3e). This is reminiscent of the condition seen in Latenivenatrix (van der Reest and Currie, 2017). In lateral view, this surface faces slightly dorsally and extends posteriorly as far as the postzygapophysis of S5 (Fig. 5b). Additionally, the posterior surface of S5 is slightly offset ventrally relative to the intercentral junction of S4 and S5 as in other theropod dinosaurs (Brusatte et al., 2012).

The neural arch is fused to the centrum in all sacrals, and the neural spines are fused into a single apron as in most other theropods (Barsbold, 1974; Barsbold et al., 2000; Brochu, 2003; Xu et al., 2006) (Fig. 5a). In contrast, separated sacral neural spines are documented in the unenlagiid Buitreraptor (Gianechini et al., 2018), and partially fused sacral neural spines are separated by fenestrae in Rahonavis (Forster et al., 1998; Forster et al., 2020). Anteriorly, a pair of spinoprezygapophyseal laminae extends from the neural spine and demarcates a spinoprezygapophyseal fossa in S1. However, because of serious erosion, little is known about the prezygapophyses themselves. The zygapophyses of the remaining sacrals have been reduced and are likely to have been fused. In dorsal view, the dorsal surface of the neural arch lateral to the neural spine appears moderately concave, but the lateral surfaces of the neural spine are flat (Fig. 5a,b). The neural arches are mostly transversely expanded at the anteroposterior midpoint, where a pair of highly eroded transverse processes project laterally (Fig. 5a,c). The right transverse process of S2 is partially preserved, extending slightly dorsally while terminating in a dorsoventral expansion for the sacral rib (Fig. 5a,b). No sacral ribs are preserved in association with the sacrum due to erosion; thus, whether the sacral ribs and vertebrae are fused remains unclear.

Only the left postzygapophysis is well-preserved in S5 (Fig. 5e). It projects posterodorsally and terminates at nearly the height of the posterior surface of the centrum, as mentioned above. The postzygapophyseal articular surface is small, and it faces laterally more than ventrally (Fig. 5e). Lateral to the base of the postzygapophysis, the neural arch is excavated by a deep, posterodorsal facing foramen on each side such that the neural arch is demarcated by a deep V-shaped notch that is located between the transverse process and the postzygapophysis when viewed dorsally (Fig. 5a). The medial aspect of the postzygapophysis connects with the neural spine proximally via the spinopostzygapophyseal lamina, and the left and right spinopostzygapophyseal laminae together demarcate a deep and elongated spinopostzygapophyseal fossa (Fig. 5a). The presence of both spinoprezygapophyseal and spinopostzygapophyseal fossae suggests that the five sacral vertebrae appear to constitute the synsacrum of LH PV39. A pair of very small posteriorly facing articular surfaces, presumably representing the hyposphenes, is located ventral to the postzygapophysis and above the neural canal (Fig. 5e). The surface between the hyposphenes is flat and faces posteroventrally. Immediately ventral to the possible hyposphenes, the extremely dorsoventrally compressed neural canal is visible, having a transverse width of 6.25 mm and a depth of 1.65 mm.

Caudals (Fig. 5h–k)

Although the total number of the caudal vertebrae of LH PV39 is unknown, four of them are preserved in sequence. Comparisons with more complete troodontid caudal series suggest that they are likely to fall somewhere within the range of Ca10-20 (Currie and Dong, 2001; Xu et al., 2002; Gao et al., 2012; Tsuihiji et al., 2014; Shen et al., 2017a; Shen et al., 2017b; Xu et al., 2017), but their exact positions are difficult to determine. Therefore, they are referred to, from anterior to posterior, as CaA, CaB, CaC, and CaD, respectively.

The centra are dorsoventrally low, approximately three times longer than they are deep (Fig. 5h–k; Table 1). The transverse processes have been reduced to weakly developed nubbins on the lateral surfaces and are difficult to discern. The neural arches and centra are completely fused in all caudals, and there is no sign of a neurocentral suture between the two components (Fig. 5h–k). The amphiplatyan centra are rod-like; all intercentral articular surfaces are only slightly concave. In lateral view, there is no offset between the anterior and posterior articular surfaces and the dorsal and ventral margins of the vertebra are gently concave at the midpoint (Fig. 5h), as is the case in most other dinosaurs (Norell et al., 2009; Averianov and Sues, 2016). Although the rough surfaces suggest serious weathering of the caudals, the lateral surfaces lack pneumatic foramina as in many other theropods, including Allosaurus (Madsen, 1976), Fukuivenator (Azuma et al., 2016), Fukuiraptor (Azuma and Currie, 2000), Anchiornis (Hu et al., 2009; Pei et al., 2017a), Sinusonasus (Xu and Wang, 2004), Sinornithoides (Currie and Dong, 2001), Zanabazar (Norell et al., 2009), Mei (Xu and Norell, 2004; Gao et al., 2012), Almas (Pei et al., 2017b), Rahonavis (Forster et al., 1998), Deinonychus (Ostrom, 1969), Microraptor (Xu et al., 2000; Xu et al., 2003; Pei et al., 2014), Velociraptor (Norell and Makovicky, 1999), Buitreraptor (Gianechini et al., 2018), and Archaeopteryx (Wellnhofer, 2009; Rauhut et al., 2018). However, this is in contrast to cases in which the foramina is present: for example, in oviraptorosaurs (Barsbold et al., 2000; Xu et al., 2007) and megaraptoran Orkoraptor (Novas et al., 2008). Depressions excavated in the lateral surfaces of the caudal centra are present in Liaoningvenator (Shen et al., 2017b), Daliansaurus (Shen et al., 2017a), Buitreraptor (Gianechini et al., 2018), and Archaeopteryx (Rauhut et al., 2018), among others. The ventral central surface is marked by a distinct shallow groove as in other coelurosaurs (Rauhut, 2003; Carpenter et al., 2005; Xu et al., 2007; Alifanov and Barsbold, 2009; Zanno et al., 2011; Averianov and Sues, 2016; Gianechini et al., 2018). A ventral extention of the posterior articular surface probably represents the facet of chevron articulation.

Neural arches are present in all preserved caudal vertebrae, which are only slightly longer than each corresponding centrum (Fig. 5h). This differs from the condition in dromaeosaurids, in which the elongate prezygapophyses are usually several times longer than the respective centrum (Ostrom, 1969; Forster et al., 1998; Norell and Makovicky, 1999; Norell and Hwang, 2004; Pei et al., 2014; Depalma et al., 2015; Motta et al., 2018). A neural spine is absent from all vertebrae; in its place is an anteroposteriorly extended groove as in all known troodontid caudals (Russell, 1969; Russell and Dong, 1993; Norell et al., 2000; Currie and Dong, 2001; Xu et al., 2002; Norell and Hwang, 2004; Xu and Norell, 2004; Norell et al., 2009; Zanno et al., 2011; Gao et al., 2012; Tsuihiji et al., 2014; Shen et al., 2017a; Hartman et al., 2019) (Fig. 5f). This groove is deeper than the one on the ventral aspect of the centrum. The zygapophyses are oriented at a low angle relative to the anteroposterior axis of the centrum (Fig. 5h,i). No complete prezygapophyseis are preserved; however, judging from their bases, the prezygapophyses are slightly more widely spaced than the postzygapophyses relative to the midline of the neural arch (Fig. 5i). There is a deeply excavated, well-defined anteriorly facing spinoprezygapopgyseal fossa between the prezygapophyses that is roofed by posteromedially converging spinoprezygapophyseal laminae. A distinct sharp ridge begins from the medial aspect of the prezygapophysis and extends posteriorly, becoming confluent with the dorsal margin of the postzygapophyses posteriorly (Fig. 5h). This ridge defines the lateral extent of the groove on the dorsal surface of the neural arch. Another ridge, the prezygopostzygapophyseal lamina, extends from the dorsolateral aspect of the prezygapophysis, and continues with the postzygapophyses to separate the lateral surface of the neural arch from its dorsal surface. A continuous prezygopostzygapophyseal lamina is also present in Zanabazar (Norell et al., 2009), Sinornithoides (Currie and Dong, 2001), Liaoningvenator (Shen et al., 2017b), Gobivenator (Tsuihiji et al., 2014), Daliansaurus (Shen et al., 2017a), and many dromaeosaurids (Forster et al., 1998; Gianechini et al., 2018). Unlike the condition of LH PV39, in Buitreraptor (Gianechini et al., 2018) and Fukuivenator (Azuma et al., 2016), among other theropods, the prezygopostzygapophyseal lamina is not continuous, fading away at the midsection in the middle caudals.

The postzygapophysis on the right side of CaA and CaB and on the left side of CaC are preserved with the articular surfaces extending only slightly posterior to their respective centra (Fig. 5h,i). The articular surfaces are fairly small and flat, distally tapering, and facing completely laterally as in Zanabazar (Norell et al., 2009). This suggests that the prezygapophyseal articular surfaces of the succeeding caudal would be short and face primarily medially, and the lateral bending capability within the caudal vertebrae of LH PV39 was, therefore, likely to be rather restricted. Due to erosion, other detailed morphology of the postzygapophysis remains unclear. There is no trace of a spinopostzygapophyseal fossa, a feature that distinguishes LH PV39 from the caudal vertebrae of Urbacodon (Averianov and Sues, 2016). The neural canal is oval-shaped in anterior or posterior views, and it is approximately as large as the spinoprezygapophyseal fossa (Fig. 5j,k). This is unlike the condition in Urbacodon in which both the caudal spinopre- and spinopostzygapophyseal fossae are apparently larger than the neural canal (Averianov and Sues, 2016).

The preserved fragmentary chevrons associated with CaA and CaB are flattened plate-like elements (Fig. 5i). The ventral surface of the relatively complete chevron bears a shallow and transversely broad midline groove. Owing to their poor preservation, other morphologies are difficult to describe.

Manual unguals (Fig. 5l–n)

Only two manual unguals were recovered, and both are strongly compressed mediolaterally and recurved (Fig. 5l–n). The side and digit to which each ungual belongs are difficult to determine. One nearly complete ungual is elongated and sharply pointed distally (Fig. 5l–n). Here, this ungual is interpreted as a manual ungual based on the following criteria: (i) the curvature of the ungual is approximately 120° (following the method in Ref. (Pike and Maitland, 2004)); (ii) the proximal articular surface is taller than the transverse width; (iii) the presence of a proximodorsal lip; (iv) a transverse groove is present, separating the proximal articular surface from the well-developed flexor tubercle; and (v), the proximal articular surface comprises two nearly equal facets. The outer curvature is 24.2 mm long, and only a very small portion of the distal tip is missing. Proximally, the ovoid articular surface is 5.8 mm deep and 2.3 mm wide, and is divided into two nearly equal concave facets by a prominent vertical medial ridge (Fig. 5n). In lateral view, the dorsal most point of the proximal articular surface contributes to a proximodorsal lip that projects primarily posteriorly and only slightly dorsally (Fig. 5l). The well-developed flexor tubercle is ventrally rugose (Fig. 5m,n) and separated from the ventral margin of the proximal articular surface by a dorsoventrally broad transverse groove that is not as deep as those seen in other maniraptorans (Novas et al., 2005; Bell et al., 2015; Hartman et al., 2019). On each side, a distinct groove is present for the vascular supply to the keratinous sheath. The groove on the left side extends from the distal extremity of the element, becoming shallower as it extends proximally to a position between the distalmost extent of the flexor tubercle and the proximal articular surface. Although the groove on the right side is roughly as deep as the medial one, it proximally terminates anteriorly to the flexor tubercle. The other preserved ungual is preserved in insufficient detail to merit description as most of its distal and proximal ends have been broken off. Nevertheless, it can be identified as a manual ungual based on its size, which is comparable to the relatively complete ungual.