4.1.1. 2D counts and measurements

A total of 191 elements were deemed to be in a suitable preservation state for a 2D quantitative study realized on SEM (Scanning Electron Microscope) pictures. Variables included the following counts (Fig. 2A):

• Number of triads (Triads_nb): estimation of the number of triads, based on the presence of the pair of outer and inner denticles;
• Number of median denticles on the spindle area (Med_dent_nb);
• Number of lateral denticle on the blade area (Lat_blade_dent);
• Number of median denticle on the blade area (Med_blade_dent).

Among these counts, the number of lateral and median denticles on the blade area were designed to characterize the helmsi morphotype, whereas the number of median denticles on the spindle area relative to the number of triads should characterized the mawsonae morphotype.

Furthermore, several 2D quantitative measurements were taken on 2D pictures of the upper (occlusal) surface of these 191 specimens using an image analysis device (Nikon NIS Elements software), in to describe the general size and shape of the elements (Fig. 2B):

• Maximum Length (MaxLength): maximum antero-posterior length of the element;
• Square root of the area of the 2D surface of the element in occlusal view (sqrtArea);
• Square root of the area of the spindle surface in occlusal view (sqrtSpinArea);
• Maximum Width (MaxWidth): largest inner-outer distance of the element;
• D1: distance between the inner and outer denticle of the first triad.

For 133 of these elements, measurements were taken on MEB pictures. A subset of 58 elements was µCT-scanned and measurements were taken on 2D snapshots of the 3D surface oriented with the occlusal surface up as for MEB pictures. The total sampling included 95 Ic. alt. alternatus s.s., 36 Ic. alt. helmsi, 17 Ic. alt. mawsonae, and 42 “small” specimens. One specimen displayed characters typical of both, Ic. alt. helmsi (most posterior denticle shifted towards the inner line of denticles) and of Ic. alt. mawsonae (absence of expressed first median denticle, md1) (Fig. 3D). Since the anterior was broken, it was discarded for most analyses, but was otherwise included in the mawsonae group.

4.1.2. Acquisition and extraction of 3D surfaces

A subset of 58 elements were glued on a toothpick and scanned at a cubic voxel resolution of ~1 µm using Phoenix Nanotom S microtomograph (μCT) on the AniRA-ImmOs platform of the SFR Biosciences, Ecole Normale Supérieure, Lyon (France). The scanning parameters were as follow: 100 kV, 70 µA, 3000 projections at 360° with no filter. The surface of the element was extracted semi-automatically using the thresholding tool in Avizo (v. 9.1—Visualization Science Group, FEI Company).

These elements were selected to document the morphological variation over a broad range of ontogenetic stages (from two to eight triads) and across subspecies. Most corresponded to alternatus specimens (N = 29), but the sampling included the helmsi (N = 11) and mawsonae (N = 5) subspecies. One specimen displayed characters typical of both, helmsi-like with the most posterior denticle shifted towards the inner row of denticles and of mawsonae with absence of expressed first median denticle, md1; it was depicted as “hybrid” but attributed to the subspecies mawsonae in statistical analyses (Fig. 3D). In the 2D analyses, “small” specimens were those with three or less triads (N = 12). Specimens with little or no damage were preferentially selected, but several broken elements were included because of their interest in documenting the morphological variation. Three specimens were anteriorly broken, hindering the measurements of their length.

All the specimens are housed in the collections of the University of Montpellier. The reconstructed 3D surfaces of the specimens illustrated on Fig. 2 and Fig. 3 (collection numbers UM BUS 031 to UM BUS 045) have been deposited in MorphoMuseuM (Girard et al. 2022).

4.1.3. Length, width and 3D landmarks

3D surfaces of the left elements, spur to the right when the element is seen in occlusal view), were subjected to a mirror transformation and measured as right elements. Length, measured as the greatest antero-posterior distance, and width, the largest inner-outer dimension, were manually measured on the 3D surfaces using Avizo. Fifteen 3D landmarks (Suppl. material 1: Table S1, Fig. 2C) were manually positioned on the surface of each element using IDAV Landmark Editor v.3.6 (Wiley et al. 2005). They document the “core” area of the element, including the blade and first two triads: being located on the first parts to develop along the ontogeny, these landmarks can be identified on all specimens including small ones. They corresponded to the tip of the cusp, the most posterior denticle (mpd) following the cusp, the tip of the inner, outer and median denticles of the first two triads, as well as the valleys between the median and lateral denticles. The pit was also documented. Defined as the deepest point of the basal cavity, its position was difficult to assess in a comparable manner in all elements, because of the occurrence of more posterior secondary “pits” in some elements. The “pit” landmark was thus taken as the deepest point in the vicinity of the first median denticle (md1).

Correlations involving counts were assessed using Kendall’s rank order test; correlations between numerical variables were assessed using the Pearson correlation coefficient. Differences between the alternatus, helmsi, mawsonae, and small groups were tested using non-parametric Kruskal-Wallis tests and pairwise Wilcoxon comparisons.

Linear models using Maximum Length as the independent variable were used to assess relationships of the different variables along ontogenetic growth, and to provide size-corrected residuals. Models including Maximum Length and group as factor were also performed. They allowed to test the significance of the differences between groups while taking size into account; the interaction between both factors indicated whether slopes were similar in the different groups.

All these tests were performed using R (R Core Team 2017).

4.3.1. Length and width

The relationship between element length and width was assessed using a linear regression. Since width was available for all elements, the relationship length ~ width was used to interpolate length values for the three anteriorly broken specimens.

4.3.2. Characterization of the element geometry using interlandmark distances

In order to document the growth of the different parts of the elements, inter-landmark distances were calculated from the landmark coordinates. The following distances were considered:

• dlat1, distance between the inner and outer denticles of the 1 ^st triad [= d(id1, od1)]. This distance is the equivalent in 3D of the distance D1 in 2D, but incorporates differences in height between the two lateral denticles of the triad;
• dlat2, distance between the inner and outer denticles of the 2 ^nd triad [= d(id2, od2)]; this is the equivalent for the second triad of dlat1;
• d(cusp, mpd), distance between the cusp and the most posterior denticle following it;
• d(mpd, md1), distance from mpd to the median denticle of the 1 ^st triad md1;
• d(mpd, id1) and d(mpd, od1), distances from mvd to the inner and outer denticles of the 1 ^st triad, respectively.

How these distances varied along the ontogeny was investigated using linear regression of each distance vs element length. Interpolated values were used for visual representations but were not included in the calculation of the regressions.

4.3.3. Denticle height and pit depth: estimation by the triangle geometry

In order to estimate the height of the denticles above the valleys, geometric properties of the triangle were used. For any triangle, the semi-perimeter (p) and the area (S) can be derived can be calculated from the sides of the triangle a, b, and c, following to the two equations p = (a + b + c) / 2 and S = sqrt[p (p-a) (p-b) (p-c)]. The altitude h at the summit A (opposed to side a) can then be obtained as h = 2S/a.

According to these formulas, the height of md1 can be approximated as the altitude of a triangle defined by the tip of the denticle (md1) and the two valleys surrounding it along a same line (iv1 and ov2, or alternatively ov1 and iv2) (Suppl. material 3: Fig. S1), the sides of the triangles being calculated as the corresponding inter-landmark distances. The height of md1 was thus calculated using the triangles md1-iv1-od2 and md1-od1-iv2 and assessed as the mean of both estimates. A similar procedure allowed the depth of the pit to be estimated from the same valley points.

The height of the outer and inner denticles were assessed using the valleys along the same line: h-id1, based on the triangle id1-iv1-ov2; h-od1 based on od1-ov1-iv2; h-id2 based on id2-iv2-ov1, and h-od2 based on od2-ov2-iv1.

The relationship of denticle height and pit depth with element length was assessed using linear regression. Linear models including, groups and their interaction were further investigated.

4.3.4 3D geometric morphometrics

The configurations of the fifteen 3D landmarks were superimposed using a generalized Procrustes analysis (GPA) standardizing size, position, and orientation while retaining the geometric relationships between specimens (Rohlf and Slice 1990). The aligned coordinates constituted the shape variables for further analyses. The size of the landmark configuration was estimated by the centroid size where CS: square root of the sum of squared distances from the landmarks and semi-landmarks to the centroid).

Size-related variations in shape and differences between groups (alternatus, helmsi, mawsonae and “small” groups were investigated using Procrustes ANOVA. With this approach, the Procrustes distances among specimens are used to quantify the components of shape variation, which are statistically evaluated via permutation, here, 9999 permutations (Adams and Otarola-Castillo 2013). The allometric relationship was visualized as the common allometric component (CAC) derived from an analysis of the aligned coordinates vs. size (Adams et al. 2013). Procrustes ANOVA were also used to assess models including size, groups, and their interaction as factors.

The Procrustes superimposition, allometric analysis, and Procrustes ANOVA were performed using the R package geomorph (Adams and Otarola-Castillo 2013).

The 2D sampling included elements with two to eight triads (Fig. 4A). Maximum length increased with the number of triads (Kendall’s Tau = 0.7928004, p < 2.2e-16). Groups were different for their maximum length (Kruskal-Wallis p < 2.2e-16), taken as a proxy of size. The subspecies alternatus covered a large size range whereas helmsi and mawsonae were both characterized by a similar large size (Fig. 4B, Table 1).

Figure 4. 

Maximum length of Icriodus elements along triad numbers (A) and among the subspecies (B) represented as dot plots. Each dot corresponds to one observation (specimen); dots within a given bin width are stacked. Here, bin width = 30.

All variables, being counts or numeric measurements, were significantly related together (Fig. 5, Suppl. material 2: Table S2). Continuous measurements were particularly highly correlated with each other (e.g. Fig. 6), due to an overall increase in all dimensions when triad number increases. The distance between the inner and outer denticle of the first triad, D1, was the less strongly correlated to the other measurements. Among counts, the number of lateral denticles on the blade was the less strongly associated with other variables. One noticeable exception emerged in this pattern: the number of median denticle on the blade was significantly related with the other variables, but it decreased with triad number and overall size.

Figure 5. 

Correlogram of the relationship between each pair of counts and numeric variables in the Icriodus dataset. Size of the circle is proportional to the strength of the correlation, color varies with the value of the correlation. Kendall’s tau correlation coefficient has been used for the representation. Variables are ordered according to a hierarchical clustering of the correlations.

Figure 6. 

Relationship between maximum length and the square root of total area across the four groups of Icriodus. The dotted line represents the linear correlation between both variables (Pearson correlation coefficient = 0.9598, p-value < 2.2e-16) .

The variables specifically designed to capture the characteristics of the subspecies indeed showed significant differences. The number of median denticles on the spindle area of course increased with triad number, but to a lesser degree for mawsonae (Fig. 7A). As a consequence, the ratio of the median denticles number vs triad number differed between groups (Table 1).

Figure 7. 

Number of median denticles on the spindle area (A), of median blade denticles (B) and of lateral blade denticles (C) along triad numbers in the four groups. The size of the symbols is proportional to the number of observations (N).

The number of median blade denticles decreases with triad numbers (Fig. 7B), but concomitantly, the number of lateral blade denticles tends to increase, particularly in helmsi (Fig. 7C). As a consequence, the number of lateral blade denticles, differed between helmsi and the other groups (Table 1). In contrast, the subspecies did not differ in the numeric variables when the effect of size increase was taken out by considering residuals vs. maximum length (Table 1).

The homogeneity of the subspecies when considering their general aspect was confirmed using linear models including maximum length and groups as factors (Table 2). While maximum length always displayed a highly significant effect, the group effect was significant only for √Spindle Area and maximum width, with no significant interaction showing that the four groups shared similar slopes of variation with maximum length. However, when excluding the “small” group, the group effect was not significant anymore, showing that the three subspecies Ic. alt. alternatus, Ic. alt. helmsi and Ic. alt. mawsonae did not differ in their overall dimensions, when taking out the general size increase.

In order to assess how morphological disparity varied along ontogeny, the variance of the different variables was assessed for each triad number. Triads 1 and 2, and 7-8, were grouped together because of insufficient data. The relative number of median denticles (Med_Dent_nb/Triads_nb) was considered together with maximum length and width, √ of the total area and spindle area, the distance between the inner and outer denticles of the first triad (D1), and the number of lateral and median denticles on the blade. To level out the effect of different scales and of size increase, variables were centered to the mean and divided by the standard deviation).

The patterns of variance through ontogeny were compared using Kendall rank order tests (Fig. 8A), showing little congruence except for maximum length, √Area, and √Spindle Area. However, all variables showed a minimum variance at the stage Triad 4 (Fig. 8B), to the exception of the number of lateral denticles on the blade, showing little variance in triads 1-2-3, but starting to increase in variance at triad 4. On the contrary, the variance in relative numbers of median denticles on the spindle was high in the first triads (1-2-3), decreased abruptly at triad 4, then re-increasing gently thereafter.

Figure 8. 

Disparity through ontogeny. The variance of the different reduced-centered variables was estimated for each triad number (triads 1-2 and 7-8 grouped). A. Correlogram showing the congruence of the pattern between variables, estimated using Kendall’s tau coefficient; B. Change in the level of variance through ontogeny. The variance of each variable has been scaled by its mean in order to represent the different lines on a single graph. Variables considered were: MaxLength: maximum length, sqrtArea: √ of the total area of the element; sqrtSpindleArea: √ of the spindle area; MaxWidth: maximum width; D1: distance between the inner and outer denticles of the first triad; LatBladeDent: number of lateral denticles on the blade; MedBladeDent: number of median denticles on the blade; relMedDent: number of median denticles on the spindle, divided by triad number.

Length vs width in the 3D dataset

When focusing on the 3D subset, length and width measured on the 3D surfaces were strongly related (Pearson’s product-moment correlation R = 0.920, p-value: < 2.2e-16), with the length increasing 1.5 faster than the length (slope of the regression: 1.509) (Suppl. material 4: Fig. S2). This regression provided interpolated length values for the three elements that have a missing anterior part.

The analysis of the heights and distances focused on the most posterior part of the element, including the first triad, because these features can be measured in almost all specimens, including small ones, and document the growth of the elements. Most variables increase with growth as indicated by increasing length (Table 3, Fig. 9). The distance between the tip of two lateral denticles of the first triad (dlat1) shows the tightest correlation with length, followed by the equivalent for the second triad (dlat2). The first outer denticle (od1) shows a more pronounced growth than the inner one (id1).

Figure 9. 

Relationship between Length (x-axis on all plots) and various inter-landmark distances, denticle height and pit depth. Dotted lines: significant linear relations. Interpolated values of length for the three elements with the anterior part missing (diamond symbol) are used for visualization but are not included in the regressions.

Three variables clearly depart from this general growth pattern. Two distances involving the most posterior denticle show no relationship with length: distance from mpd to od1 and md1. Most strikingly, the height of the first median denticle (md1) is not only stable through growth, but even decreases. Concomitantly, pit depth increases slightly faster than the height of md1. Pit depth roughly assesses the thickness of the element in the vicinity of md1, from the same valley points from which md1 height is calculated. The balance between increasing pit depth and decreasing md1 height may therefore be largely due to a filling of the valleys, ultimately leading to a md1 which is not visible anymore.

Variables can thus be associated according to the way they covary (Fig. 10). A first block of variables shows a pronounced association with length: dlat1, dlat2, pit depth, and od1 height; md1 height is also strongly associated with length, but in a negative way. The distance of the most posterior denticle (mpd) to id1 and md1 is not related to length, but is associated with other variables related to distance from mpd to the pit and to od1. The height of id1, id2 and od2 are relatively loosely related to the other variables.

Figure 10. 

Correlations between univariate variables: element length (including the three interpolated values), denticle height, pit depth, and distances between denticles. In each case, the diameter of the circle and its color is proportional to the strength of the correlation, estimated using the Pearson coefficient. Variables are order to a hierarchical clustering approach.

These analyses were complemented by linear models including length, groups, and their interaction, in order to assess whether the growth dynamics differed between alternatus, helmsi, mawsonae, and the small specimens (Table 3). For all variables which were significantly related to length with a simple regression, this relation was confirmed and no difference between groups was indicated.

The Procrustes superimposition delivered the centroid size of the configurations as another proxy of element size, which was related to element length (Pearson’s product-moment correlation R = 0.673, p-value = 1.811e-08) (Suppl. material 5: Fig. S3).

The morphometric analysis of the 3D shape showed along the first axis of the PCA on the aligned coordinates (PC1 = 35.2%), an opposition between small elements, towards negative scores, and helmsi, and especially mawsonae, tending to display high positive scores (Fig. 11). Scores on PC1 were highly correlated with element size (correlation with Length, including interpolated values: R = 0.891, p-value < 2.2e-16).

Figure 11. 

3D Geometric morphometric analysis of Icriodus alternatus shape variation; A. First two axes of a PCA on the aligned coordinates; B, C, D. Visualization of the deformation in profile and oral view; in yellow tip of the denticles, in green valley landmarks, in black the pit; B. Configuration corresponding to the minimum score along PC1; C. Shape change from the minimum (dots) to the maximum (tip of the vectors) scores along PC1; D. Configuration corresponding to the maximum score along PC1.

Therefore, the variation along PC1 mostly depicts the growth of the element. Towards negative scores corresponding to small elements, outer and inner denticles are close to and approximately at the same distance from the median axis of the element (Fig. 11B). The first median denticle is the highest, and the pit is located relatively posteriorly. Towards positive scores (Fig. 11D), in contrast, the pit is shifted towards a more anterior position. Inner and outer denticles grew in height and centrifugally from the median axis of the element; the centrifugal growth of the outer denticles being more pronounced than for the inner ones, the element acquired a torsion with an increased outer component. The most posterior denticle, maintaining its distance to the first outer and median denticles, tends to grow with an inner component. The height of the first median denticle is greatly reduced, a trend shared by the second median denticle, but to a lesser degree.

The relationship of shape with size and groups were investigated using Procrustes ANOVA. As expected, the size / shape relationship was very strong (Procrustes ANOVA, shape ~ Length including interpolated values: P = 0.0001) (Suppl. material 5: Fig. S3). When including groups as cofactors, Length remained the prominent factor, the grouping effect being close to the significance threshold (Procrustes ANOVA, shape ~ Length * Groups: P_Length = 0.0001, P_Groups = 0.0627, P_interaction = 0.3435). This difference between groups disappeared when small elements were grouped with alternatus (P_Length = 0.0001, P_Groups = 0.3516, P_interaction = 0.3900).

The allometric size/shape relationship can be visualized using scores along the Common Allometric Component, which highly resembles scores on PC1 (Pearson’s product-moment correlation R = 0.998, p-value < 2.2e-16), underlining the importance of the allometric signal in the total shape variation. Accordingly, shape deformation associated with allometry (Fig. 12) resembles those expressed along PC1. Small specimens are characterized by shallow inner and outer denticles, organized symmetrically on both sides of the median axis of the element; the pit is shallow as well. Large elements are characterized by a deep pit, as well as elevated inner and outer denticles, the outer one being more centrifugal relative to the median axis of the element than the inner one.

Figure 12. 

Allometric shape variation in Icriodus alternatus; A. Relationship between the total length of the element and the Common Allometric Component, based on the aligned coordinates of the posterior part of the element; B, C, D. Visualization of the deformation in profile and oral view, corresponding to extreme size values; in yellow tip of the denticles, in green valley landmarks, in black the pit; B. Configuration corresponding to the minimum length; C. Shape change from the minimum (dots) to the maximum (tip of the vectors) length; D. Configuration corresponding to the maximum length.

The 3D analysis was focused on the posterior, ontogenetically oldest part of the Icriodontan element. Results showed that independently of the addition of new triads along ontogeny, the geometry of this posterior part changed deeply along growth, the main components being: (1) a growth of the inner and outer first denticles with both, a vertical and lateral component, the centrifugal growth being more pronounced for the outer denticles; (2) a growth of the cusp towards a more posterior direction, increasing its distance from the first triad; (3) a shift of the pit towards a more anterior position, together with an increase in the thickness of the element between the pit and the valleys surrounding the first median denticle; (4) a decrease in height of the first median denticle, up to its disappearance in most extreme cases. (5) In contrast, the zone comprising the most posterior denticle, the first median denticle and inner denticle, which seem to form a “core” area of the element in which the geometry, designed early during ontogeny, is little affected later on.

The subspecies Ic. alt. alternatus, Ic. alt. helmsi, and Ic. alt. mawsonae appeared to share the same ontogenetic trajectory, and once size-related variation is accounted for, they did not differ in any quantitative variable describing their general shape or the relationships between posterior denticles. The most prominent criterion distinguishing Ic. alt. helmsi and Ic. alt. mawsonae is their large size, suggesting that actually, these subspecies simply represent end-member geometries achieved at late growth.

The Ic. alt. helmsi subspecies has been characterized by the alignment of the most posterior denticle with the inner row of denticles. This morphology is achieved due to pronounced centrifugal growth of the outer denticles. Combined with the invariant relationships between the most posterior denticle and the first inner one, this trend tends to twist the shape of the large elements, up to orienting the most posterior denticle with the inner row in the most extreme cases.

As for the Ic. alt. mawsonae subspecies, it corresponds to an end-member of the trend of decreasing height of the first median denticles along ontogeny. Since the removal of material from a denticle seems unlikely, this decreasing height, together with the increasing thickness of the element at the vertical of the first median denticle expressed by the increase in pit depth, suggest a progressive filling of the initial deep valleys. Without a concomitant growth of the median denticle, this leads to a progressive reduction, up to its disappearance. In agreement with this interpretation of the subspecies as part of a morphological continuum including all Ic. alternatus subspecies, several specimens first identified as Ic. alt. mawsonae displayed “clear” traces of the first median denticle on the 3D scans. Furthermore, at least one specimen displayed the diagnostic features of both Ic. alt. helmsi and Ic. alt. mawsonae.

As a consequence, the described subspecies appear to belong to a single, homogeneous taxonomic and evolutionary unit, corresponding to the species Ic. alternatus. Although they can be seen as a way of describing an extensive morphological variation, the use of the “subspecies” concept in this context is misleading. For modern organisms, this notion corresponds to geographically isolated pools (Mayr 1963), implying some degree of gene flow breakdown. This is clearly not the case for Ic. alt. alternatus, Ic. alt. helmsi and Ic. alt. mawsonae, and the terminology of “subspecies” should be avoided since all can be found in the same time and place, indicating shared gene pool simply corresponding to Icriodus alternatus. If ever, the term “morphotype” should be preferred to describe this morphological variation, avoiding the risk to count such units in diversity analyses.

Variation in the oldest, posterior part of the element: a general feature of the genus Icriodus

The posterior part of the platform Icriodontan element is the first to be formed in ontogeny. As such, it is exposed to remodeling during all subsequent growth, consequently being the most variable zone in different Icriodus species. A morphotype differing in the expression of the denticles in the posterior area of the spindle middle row, hence similar to Ic. alt. mawsonae, has been described in Icriodus subterminus (Middle-Late Devonian) (Narkiewicz and Bultynck 2010). In another species, Icriodus sp. cf Ic. brevis (van den Boogaard 1983), the posterior denticles of the spindle, being progressively covered by new lamellae, become progressively less sharply separated from each other. The trend culminates for the three posteriormost denticles of the median row that become almost completely fused into a crest.

The blade, the other part of the element that composes the initial growth stage, displays similar trends of denticle progressive disappearance or fusion, as shown by the progressive diminution of blade denticle number along growth in Ic. alternatus (Dreesen and Houllenbergs 1980).

Slight discrepancies exist in the temporal extension of the “subspecies” Ic. alt. alternatus, Ic. alt. helmsi and Ic. alt. mawsonae, apparently arguing for them being distinct evolutionary units. Icriodus alternatus as a whole appeared during a period marked by a succession of environmental perturbations. The mass extinction marking the Frasnian – Famennian boundary was the culmination of the Upper Kellwasser event materialized by anoxic deposits in many marine environments: it was associated with a pronounced temperature decrease and sea-level fall biosphere (Joachimski and Buggisch 2002, Girard and Renaud 2007). This event was preceded, ca. 1 myrs before, by the Lower Kellwasser event, episode of similar nature but of lesser impact on the biosphere (Joachimski and Buggisch 2002, Girard and Renaud 2007). A temperature decrease associated with a sea-level shallowing occurred towards the end of the LKE. The typical Ic. alt. alternatus form, together with Ic. alt. helmsi, appeared just after the LKE, whereas mawsonae appeared shortly before the UKE. Reversely, Ic. alt. helmsi and Ic. alt. mawsonae went extinct slightly before the typical Ic. alt. alternatus form during the early Famennian. Trends of size variation have been documented during this time period (Girard and Renaud 1996) that can explain these discrepancies. Icriodus alternatus displayed an increase in size in the interval between both Kellwasser events, a trend that should have promoted the apparition of Ic. alt. mawsonae, characterized by the largest size. Reversely, a decrease in size shortly before the species extinction might explain the disappearance of Ic. alt. helmsi and Ic. alt. mawsonae large-sized forms slightly before the final disappearance of the species. This suggests that Icriodus alternatus responded to environmental variations by shifts along a conserved ontogenetic trajectory, an interpretation supported by the allometric relationship between size and shape documented across the UKW and the Frasnian – Famennian crisis (Renaud and Girard 1999).

The ontogenetic pattern of Icriodus alternatus I elements further displayed heterogeneous disparity, with minimum morphological variance at the stage of four triads. This pattern is reminiscent of the “hourglass” developmental model (Irie and Kuratani 2014), stating that the most conserved embryonic stage is not the earliest but a mid-embryonic period called the ‘phylotypic period’. This pattern has been interpreted as due to a maximum of molecular integration leading to high developmental canalization at the stage when common anatomical features of the basic body plan are defined. Regarding Icriodontan elements, this would correspond to relatively poorly constrained initial stages of development, when the earliest triads are added to the blade, converging towards the typical phenotype of the species, before that continuous growth and remodeling generate extensive morphological variation when late triads are added. Indeed, surprisingly high phenotypic variation has been found early during embryonic tooth development (Hayden et al. 2020) despite canalized adult phenotypes. Such early variation may be favored because the tooth-like structures are not yet functional. Later on, functioning may itself contribute to generate phenotypic variation in feeding structures. The morphology of Icriodontan elements suggests that the denticles were interlocked during occlusion; strong canalization would be required so that the opposite denticles would match with each other. The movement of occlusion seems to have incorporated some rotational dynamics, not only along the antero-posterior axis of the element, but also along the inner-outer axis (Suttner et al. 2017). Possibly, the addition of lamellae during late growth was influenced by the physical pressures exerted during occlusion, thus fostering the increasing asymmetry between the inner and outer denticles of the element. The “hourglass” pattern of morphological disparity along ontogeny thus suggests a shift from relatively loose developmental constraints to a pattern of growth modulated by functional constraints during occlusion. As recently suggested for the conodont genera Polygnathus (Renaud et al. 2021) and Palmatolepis (Renaud, Girard and Dufour in press), patterns of morphological variance, and not only shape itself, may be enlightening for deciphering the functional constraints related to occlusion on the platform element geometry.