Firstly, early seedling development was monitored for four ant plant species: Anthorrhiza bracteosa C.R. Huxley & Jebb, Hydnophytum moseleyanum Becc., Myrmecodia tuberosa Jack, and Myrmephytum beccarii Elmer. The early steps of seedling development after germination were similar among the species. First, a thick hypocotyl appeared from the seed (Figure 1A) and an extensive hypocotyl thickening occurred without elongation of the main root (Figures 1B–E). Main and adventitious root elongation followed before extension of cotyledons (Figure 1F). Extension of cotyledons occurred 20-25 days after sowing. Similarly, unlike usual angiosperm seedlings, the Rubiaceae ant plant seedlings paused for a long time after cotyledon expansion for shoot apical meristem activity (Figures 1G–I). The first pair of foliage leaves become to be enough large to be seen by naked eye more than three months after sowing. Until it they therefore only developed a hypocotyl to make domatia (Figure 1) as described by Huxley (1978), assimilates obtained from expanded cotyledons are mostly consumed for development of hypocotyls. It is known that development of sink capacity of hypocotyls in radish is dependent on an ability of sucrose storage governed by sucrose synthase (Usuda et al., 1999). It is plausible that the very strong sink capacity in the swelling hypocotyl/domatium in Rubiaceae ant plants might be also regulated by such sucrose storage ability. Physiological examination from this view point is a future issue.

Interestingly, negative gravitropism was not recognized in the thickened part of hypocotyls, whereas it was observed in the thin, apical part of hypocotyls (Figure 1H). After the formation of the pore, we observed that the cavity was already established, suggesting that cavity formation precedes the opening of the pore.

As mentioned in the Introduction, our past observations of pore formation in the ant plants’ native habitat urged us to examine whether gravity indeed affected the position of the pore. As a preliminary test, Myrmecodia tuberosa seedlings with their natural positions were examined for the pore position seven months after sowing. All examined seedlings had pores at the basal part of the hypocotyl; 23 seedlings had the pore on the lower side whereas eight seedlings had it on the lateral side. Therefore, it is strongly suggested that pore position is determined by two factors: positional cue (basal position of the hypocotyl) and gravity.

To examine the influence of gravity, two weeks after the germination the seedlings of four species were laid sideways by fixing the position of the Jiffy-7 base to be tilted at a 90-degree angle (Table 1; Figures 1G–I). Due to the limited number of available seeds, other conditions for gravity stimulus timing were not tested. After 14-22 weeks, the position of the pore was checked. As shown in Table 1, all seedlings formed pores at the base of the hypocotyl, indicating that the potential for pore formation is restricted to the basal part of the hypocotyl at the seedling stage. Moreover, almost all seedlings (see Table 1) showed pore formation on the lower side of the hypocotyl basal region, clearly indicating that the pore formation site is also dependent of gravity. While the bilateral symmetry axis made by the pair of cotyledons was also considered for the positioning of pores in Hydnophytum moseleyanum, no correlation was recognized between the plane of the cotyledons and the pore position (three individuals made pore parallel to the cotyledon plane and three made perpendicular to the cotyledon plane). Therefore, pore site determination was concluded to be developmentally (at the base of hypocotyl) and environmentally (by direction of gravity) controlled.

This conclusion is reasonable considering the ecological aspects of the domatia. In the native habitat in Papua, nearly all seedlings were attached on the surface or cracks of tree trunks in a diagonal manner (Supplemental Figure 1). In this orientation, if the pore formed in the upper side of the domatia, raindrops would penetrate into the cavity and cause severe damage for tissues under construction of cavities via some infections of harmful fungi and bacteria. Therefore, the pore location on the loser side is beneficial to ant plants.

Once the location of the pore was determined, we investigated how the cavity was formed using micro CT observations. Preliminary examination showed that domatia tissue was easily observed without any staining or pre-treatment; thus, we observed the cavity of species from two major genera with past reports on their cavity shape. The Myrmecodia and Hydnophytum species had the earliest pore formation among the four species examined (Table 1), and the initial cavity shape differed between them. At 7.5 months after sowing, M. tuberosa showed very complex, branched cavities in the hypocotyl whereas H. moseleyanum had a single, large cavity (Figure 2). This is similar to past reports by Huxley (1978) which state that “the shape of the first cavity does not change markedly except in some Myrmecodia spp”. However, Huxley (1978) described the Myrmecodia cavity apex as bifid, which differs from the present observations of a branching pattern (Figure 2). In many cases, the first cavity connected to the pore had an upside-down U-shape (Figures 2A, B). In some cases, the upside-down U-shaped cavity had short or long extensions at the apex, forming a Y-shaped cavity (Figures 2A, B). Therefore, it appears that the upside-down U-shape is a basic form of the first internal cavity in M. tuberosa. Recognition of such variation is a benefit of the CT observations that was not easy in the past due to the difficulty of section-based studies to grasp the 3D shape of the cavity.

In addition, a unique character was also recognized for the M. tuberosa internal cavity. As seen in Figures 2A, B, an additional and independent cavity was formed at the apical region of the hypocotyl in M. tuberosa. The apical cavity was found to be isolated from the basal cavity which was connected by a pore and had a complicated looped shape with branching. This should be a secondary cavity described before (e.g., Jebb, 1991). In the past the role and fate of the first cavity in the mature plants was in a debate. Future time-course studies of CT-based cavities will fix the matters.

We focused on the Hydonophytum species cavity formation, because this species consistently formed the same single, large cavity in the hypocotyls. From serial observations we found several new features of the cavity that were not previously described.

Firstly, the cavity shape was always straight. Among the 14 seedlings examined, all the cavities were straight and parallel to the longitudinal axis of the hypocotyl, independent of the seedling orientation, and irrespective of whether the hypocotyls stood straight or obliquely (Figure 3). This observation indicated that the cavity shape is determined innately and is independent of gravity. The cavity extended to the apical and basal regions of the hypocotyl and ended apically beneath the shoot apex, remaining several millimeters away (Figure 3).

Secondly, initiation of the cavity started from the center region in terms of the diameter and from the mid region of the hypocotyl in terms of the longitudinal axis (Figure 4). Initially, small punctate pores (Figure 4A), which appeared to be a symptom of cell death, were formed and made the tissue sponge-like. Subsequently, they appeared to fuse into a large cavity (Figure 4B). Surrounding tissues of the cavity were then occupied with continuous formation of rough spongy structures that were then fused into the central cavity. The cavity formation expanded longitudinally and a single, large, and straight cavity was made (Figure 4C).

Thirdly, pore formation did not occur until the above-mentioned inside first cavity reached to the bottom (Figure 4C). This is contradictory to the past understanding that of cavity and pore forming in a single step after determination of the phellogens-surrounding region (Huxley, 1978). Instead, it was found that pore formation only started after cavity formation was completed. As shown in Figures 3 and 5, the cavity is always straight and parallel to the longitudinal axis of the hypocotyl, and the short tunnel that connects the basal most end of the cavity and the pore starts to form from the inner side (Figure 5A). The short tunnel then extends to the surface of the hypocotyl to open the pore (Figure 5B). Considering the fact that the inside cavity formation starts from the mid region and extends in both apical and basal directions, this unidirectional extension of the short tunnel appears to be another unique feature. The common feature between the cavity and the short tunnel for the pore is that they both arise from degradation of sponge-like tissue (Figures 4, 5). In some cases, the pore was covered with thin tissue (peel), (Figure 5B) but often this structure had naturally fallen off after shrinkage.

Fourthly, the inside cavity became completely empty even before pore formation as shown in Figures 2, 4 and 5. This observation indicates that the dead cells simply thrived. This is also inconsistent with the past statement that inner dead cells are “normally removed by the ants” (Huxley, 1978).? To elucidate how these dead cells were removed, the complete cavity region including the peel of the pore was cross-sectioned (Figure 6). As a result, it was found that the surface of the cavity and the short tunnel are both covered with thin, dead, multicellular layers (Figures 6A, B). This was also true for the peel of the pore (Figure 6C), indicating that dead cells were reduced to very thin tissue. Still in an exceptional case, a small fragment of white, sponge-like tissue remained attached to the peel and was found to be a low-density cluster of dead cells (Figure 6D). Therefore, majority of dead cells are dried up and lost in the space, but some remaining dead tissues might be removed by ants in nature. Clear cell files were not seen in majority of the parenchymatous tissue of the domatia (Figure 6A), except on the surface where two to three cell files were recognized parallel to the surface layer (Figure 6B). In some cases, crystals of calcium oxalate, that are also seen in parenchymatous tissue in living tissue as idioblasts, were seen among the dead cells in the peel (Figure 6D).

The present study revealed many hidden processes of cavity formation in ant plants that differ from past descriptions. The inconsistency appears to be derived from the methodology used for observations. In the past, time-course analysis was impossible because all the processes, except for the final step of peel formation and its removal, occur inside of thick tissue. Moreover, previous studies on cavity formation made comparisons of sections from different individuals of different ages; therefore, there was room for speculation regarding the connection between the different samples. However, as revealed in this study, all the steps of the first cavity formation is done deep inside tissue and no symptom is observed from outside until the pore opening. In addition, most past studies were carried out by using a limited number of wild individuals; therefore, it was difficult to capture each developmental step in high resolution.

In this study non-invasive CT scanning was used to solve the above difficulties in combination with cultivation under controlled conditions without association with ants. As a result, many hidden characters in the process of cavity formation could be revealed. Recently, the use of CT imaging has expanded in Plant Science; however, its use is limited (Piovesan et al., 2021). The present study demonstrates a powerful application of CT for analyzing plant morphogenesis, and it provides a rationale for further use of CT imaging by plant scientists. Importantly, no staining or pre-treatment was required for visualization of the cavities in this study, thereby enabling us to do time-course observations without damaging the seedlings.

Moreover, it was revealed that the pore position is dependent on gravity. This finding was also achieved by cultivation of seedlings under controlled conditions without ants. To date, many developmental processes have been shown to be affected by gravity response in plants (Nakamura et al., 2019); however, a suggestion of gravity-dependent positioning of the pore for symbiosis with ants is a new. Such gravity responses are often mediated by auxin signaling, which is also a major signaling pathway for controlling organ formation and morphogenesis in plants (Vanneste and Friml, 2009; Swarup and Bhosale, 2019). Moreover, in many cases auxin works as a positional cue (Teale et al., 2006; Hellmann, 2021). In plants, the most studied auxin-regulated autonomous cavity formation is aerenchyma formation in water-submerged plants such as rice (Colmer, 2003; Yamaguchi et al., 2019). For example in rice roots where extensive studies have done for mechanisms of aerenchyma formation, it was revealed that a part of AUXIN/INDOLE-3-ACETIC ACID PROTEINS (AUX/IAA) genes (IAA13) and a member of AUXIN RESPONSE FACTOR (ARF) gene (ARF19) are involved in aerenchyma formation co-acting with a LATERAL BOPUNDARY DOMAIN (LBD)-CONTAINING PROTEIN (LBD1-8) (Yamaguchi et al., 2019). Treatment with an auxin polar transport inhibitor, N-1-naphthylphthalamoc acid, was shown to decrease the aerenchyma formation in rice root that could be partially rescued by application of IAA. Thus, future physiological studies can focus on whether the gravity response, in terms of pore positioning, is mediated by auxin signaling.

Fresh seeds of Rubiaceae ant plants were obtained from STRINGEPLANTS (Nagoya, Aichi, Japan). Seeds were sown on Jiffy-7 (Sakata Seed Corporation, Yokohama, Japan) fixed on rockwool (Toyobo) moistened with tap water. Cultivation was done at 23°C, under continuous illumination with light-emitting diode or white fluorescent light at 40-60 μmoles m^-2 s^-1 as described by Tsuge et al. (1996). To examine the influence of gravity direction for the pore location determination, soon after extension of cotyledons, we changed the position of seedlings. Namely, seedlings were laid sideways by fixing the position of the Jiffy-7 base on rockwool to be tilted at a 90-degree angle (Figures 1G–I).

Sectioning, staining and microscopic observations were carried out as described by Tsuge et al. (1996). Briefly, tissue samples were fixed with formalin-acetic acid-alcohol solution and dehydrated with ethanol series and embedded in Technovit 7100 resin (Kulzer and Co., Wehrheim, Germany), followed by 10-12 μm-thickness sectioning with a microtome (Microm HM360, Thermo Fischer Scientific, USA). Sections were stained with 0.1% (w/v) toluidine blue (Sigma) in 0.1 M phosphate buffer (pH 7.0), washed with water, dried, and mounted on glass slides with Entellan new mounting medium (EMD, Millipore, Burlington, MA, USA). Microphotography was done using a differential interference contrast microscope (DM4500, Leica, Germany).

CT imaging was done without any staining or pre-treatment of plant individuals cultivated on Jiffy-7 under 130 kV, 300 μA, and high resolution scan mode using a micro-CT scanner (Rigaku CT Lab HX; Rigaku, Japan). The voxel size of 3D-CT image was 37.4 μm. The 3D-CT images were analyzed in iso-surface rendering without applying any image filters. The obtained series of images were integrated into 3D images from where x-, y-, and z- sections were automatically reconstructed using myVGL software (Volume Graphics Co., Ltd., Nagoya, Japan).