Ca2+ Depletion from Granules Inhibits Exocytosis

The secretory compartment is characterized by low luminal pH and high Ca2+ content. Previous studies in several cell types have shown that the size of the acidic Ca2+ pool, of which secretory granules represent a major portion, could be estimated by applying first a Ca2+ionophore followed by agents that collapse acidic pH gradients. In the present study we have employed this protocol in the insulin-secreting cell line Ins-1 to determine whether the Ca2+ trapped in the secretory granules plays a role in exocytosis. The results demonstrate that a high proportion of ionophore-mobilizable Ca2+ in Ins-1 cells resides in the acidic compartment. The latter pool, however, does not significantly contribute to the [Ca2+] i changes elicited by thapsigargin and the inositol trisphosphate-producing agonist carbachol. By monitoring membrane capacitance at the single cell level or by measuring insulin release in cell populations, we show that Ca2+ mobilization from nonacidic Ca2+ pools causes a profound and long lasting increase in depolarization-induced secretion, whereas breakdown of granule pH had no significant effect. In contrast, releasing Ca2+ from the acidic pool markedly reduces secretion. It is suggested that a high Ca2+ concentration in the secretory compartment is needed to sustain optimal exocytosis.

A rise in intracellular Ca 2ϩ concentration ([Ca 2ϩ ] i ) is necessary to induce regulated secretion in most cell types (1,2). In neurons, [Ca 2ϩ ] i increases up to several hundred M are needed to trigger vesicle fusion, whereas in endocrine cells, granule exocytosis appears to require lower [Ca 2ϩ ] i rises (3)(4)(5)(6)(7). The time course of exocytosis also appears different in the two cell types. Synaptic vesicle fusion is very fast (s) and abrupt, whereas granule fusion is slower and more sustained (3,8,9).
Aside from these differences, important similarities exist between secretory vesicles and granules. Both secretory vesicles and granules contain large amounts of Ca 2ϩ ions (10 -13). The function traditionally attributed to the high Ca 2ϩ content in the secretory compartment is the packaging and processing of intravesicular content (14,15). More recently a granular localization of the type 3 InsP 3 1 receptor has been suggested, based on high resolution immunocytochemistry of pancreatic ␤-cells (16). In the exocrine pancreas, evidence has been provided suggesting that the intragranular Ca 2ϩ content is released by opening of low affinity InsP 3 receptors (17). These conclusions, however, have recently been challenged (18,19), and the role of granular Ca 2ϩ remains elusive. Another line of evidence suggesting that intragranular Ca 2ϩ is implicated in secretion comes from the recent identification of an acidic Ca 2ϩ -binding protein, granule lattice Protein 1 (Grl1p), in dense core secretory granules of Tetrahymena thermophila that appears essential for regulated secretion (20).
Another common feature between secretory vesicles and granules is their low luminal pH. They share this characteristic with the lysosomal/endosomal compartment and the trans-Golgi network (21,22). Indeed, the low pH of the lumen has proven a reliable means for determining the Ca 2ϩ content of the so-called "acidic Ca 2ϩ pool." Since ionophores such as ionomycin or A23187 are largely ineffective in transporting Ca 2ϩ from an acidic environment (28), the pH gradient between lumen and cytosol must be collapsed before they can effectively release the Ca 2ϩ content of this pool into the cytoplasm (23)(24)(25)(26).
The aim of the present study was to establish whether the Ca 2ϩ stored within the acidic pool is important in the late steps of exocytosis. For this purpose we employed as a model system the ␤-cell line Ins-1, an insulin-secreting cell line established from a rat insulinoma that, among different lines, best retains the phenotype of ␤-cells (27)(28)(29). Among other properties, Ins-1 cells display temperature-dependent and glucose-responsive secretion and, as shown here, temperature-dependent increases in membrane capacitance upon depolarizing pulses. Therefore, they can be used as an alternative to ␤-cells for studying secretion at the single cell level. By using capacitance measurements in combination with agents that mobilize Ca 2ϩ and/or collapse intracellular pH gradients, the role of different intracellular Ca 2ϩ pools in secretion has been assessed. We here demonstrate that, although a low pH in the granules is not required for the late steps of secretion, the level of intragranular Ca 2ϩ significantly affects the secretory profile.
Ca 2ϩ Measurements in Ins-1 Cells-Cells were loaded for 30 min at 37°C with 2 M fura-2/AM as described previously (26) 2.8 mM KCl, 1 mM MgCl 2 , 2 mM CaCl 2 , 5.6 mM glucose, 10 mM HEPES, pH 7.4, at 33°C. Cells were placed on the stage of an inverted microscope equipped with a 40ϫ oil immersion objective (Zeiss, Germany) and connected to a digital video imaging system (Georgia Instruments, Roswell, GA). All experiments were performed at 31-33°C. Excitation wavelengths were set at 340 and 380 nm. Fluorescence emission at 510 Ϯ 15 nm was collected by a CCD camera, and 8 images were averaged/time point. Time series were acquired with a frame interval of 4 s, and images at both excitation wavelengths were stored on an optomagnetic disc recorder (Panasonic, Japan). All data were normalized to the first min base-line ratio.
The total content of cellular Ca 2ϩ under different conditions was assayed by atomic absorption spectrophotometry. Cells (20 ϫ 10 7 cells/ ml) were suspended in a Ca 2ϩ -free medium containing 1 mM EGTA and challenged with ionomycin (10 M) alone or in combination with monensin (10 M). Cell aliquots (10 7 ) were then centrifuged for 5 min at 14,000 rpm in Eppendorf tubes containing 100 l of sucrose 12.5% and 400 l of silicon oil. The pellet was resuspended in 1 ml of a solution containing Triton (0.02%) and NaOH (0.2 N) before measurement.
Membrane Capacitance Measurements-Unless otherwise specified, during electrophysiological recordings, cells were perfused at 31-33°C with an external solution containing 118 mM NaCl, 20 mM tetraethylammonium chloride, 5.6 mM CsCl, 1. Membrane capacitance (Cm) was measured with the "sineϩdc" mode of the "lock-in" extension of the Pulse software, based on the Lindau-Neher algorithm (30). An 800 Hz, 40 mV peak-to-peak sinusoid stimulus was applied to the DC holding potential of Ϫ80 mV. During a depolarizing pulse and 5 ms before and after the pulse, no sinewave was applied. No leak subtraction was performed on the evoked currents in the calculations used. After the whole-cell configuration was established, Cm was recorded and canceled by the automatic capacitance compensation of the EPC-9. The procedure was repeated every 180 s to prevent a possible saturation of the lock-in signal (31).
For fast capacitance changes after depolarizing pulses, it has been reported that activation of a Na ϩ current can lead to transient increases in Cm not linked to exocytosis (⌬Ct) (32). ⌬Ct contributed maximally 10% of the initial ⌬Cm and returned to zero within 100 ms after the depolarization. To reduce its contribution on ⌬Cm measurements, the first 100 ms of the trace was discarded. The capacitance values of the following 200 ms were averaged and compared with the capacitance values before the depolarization to obtain the ⌬Cm reported in Figs. 6 and 7.
The [Ca 2ϩ ] i was monitored by a photometry equipment (T. I. L. L. Photonics, Germany) controlled by the fura-2 extension of the Pulse software (HEKA) as described previously (33). Calculation of [Ca 2ϩ ] i was performed on the calibrated ratio values (360 nm/380 nm), where R min (0.86), R max (6.07), and K-factor (1.68 ϫ10 Ϫ3 ) were obtained by an internal calibration procedure. The fura-2 fluorescence, the holding current, the lock-in, and other parameters were synchronously recorded also at low resolution (3 Hz) by the X-Chart extension of the Pulse software (HEKA). Time courses of Cm as shown in Figs. 3-5 were obtained from the Cm trace, recorded at low frequency by a point-bypoint subtraction. Positive and negative values are indicative of, respectively, exocytotic and endocytotic events. The cytosolic pH was monitored by ratioing the BCECF fluorescence signal (510 nm), excited at 440 and 490 nm.
Ionomycin and monensin were prepared from stock solution in Me 2 SO (or ethanol) (0.4% final concentration) in the Ca 2ϩ -free external solution containing 5 mM EGTA to prevent cells from loading with extracellular Ca 2ϩ in the presence of ionomycin. All drugs were applied by local pressure from a wide-tipped micropipette (5-10 m) positioned close to the cell.
Insulin Secretion Studies-Superfusion experiments on Ins-1 cell suspensions were performed as described previously (34). In short, cells were brought into suspension and placed in 1-ml superfusion chambers at a density of 10 6 cells/chamber. The cells were superfused at a rate of 1 ml/min at 37°C, and test substances were introduced with the buffer. One-min fractions were collected and subjected to an insulin radioimmunoassay. For presentation, the KCl-induced stimulation of the second pulse was integrated and normalized to that of the first pulse.

[Ca 2ϩ ] i Dynamics in Ins-1 Cells-When [Ca 2ϩ
] i was monitored by fura-2 in intact cells bathed in a medium containing CaCl 2 (2 mM) and glucose (5.6 mM), approximately 50% of the cells displayed asynchronous oscillations of [Ca 2ϩ ] i , whose frequency and amplitude largely depended on the cell batch (29). In Fig. 1, the [Ca 2ϩ ] i kinetics from several individual cells were averaged, leading to a partial masking of the initial oscillations. The addition of EGTA immediately abolished these oscillations and caused a decrease in the base-line ratio, indicating that they depend on Ca 2ϩ influx. The presence of acidic Ca 2ϩ pools and their contribution to [Ca 2ϩ ] i rises were evaluated with the protocol previously employed in other cell lines (23)(24)(25)(26). In Fig. 1A, after EGTA addition, the fast-exchangeable Ca 2ϩ pool was released by the Ca 2ϩ pump inhibitor thapsigargin (Tg) (1 M). The subsequent addition of the Ca 2ϩ ionophore ionomycin (1 M) led to a small, further increase in [Ca 2ϩ ] i , indicating that in these cells the large majority of the ionomycin-sensitive pool is represented by the Tg-sensitive one. Release of the acidic Ca 2ϩ pool was then achieved by the addition of the Na ϩ /H ϩ exchanger monensin (2 M). Qualitatively similar data have been obtained by addition of the weak base chloroquine (40 M), used in place of monensin to dissipate the intraluminal pH gradients. The increase in [Ca 2ϩ ] i after monensin (or chloroquine) application requires the pretreatment with ionomycin (26). In fact, addition of either drug alone was without appreciable effect on [Ca 2ϩ ] i (data not shown). Integrating peak areas showed that the amount of Ca 2ϩ residing in acidic compartments was, on average, 51.3 Ϯ 3.6% (n ϭ 4) of total releasable Ca 2ϩ . The amount of Ca 2ϩ released from the different Ca 2ϩ pools was also assayed by atomic absorption spectrophotometry. In controls (unstimulated conditions), the total content of cellular Ca 2ϩ was estimated to be 5.7 nmol of Ca 2ϩ /mg of protein (n ϭ 3). Ionomycin alone or ionomycin and monensin together released 2.4 Ϯ 1.1 and 4.5 Ϯ 0.4 nmol of Ca 2ϩ /mg of protein, respectively (n ϭ 3). Thus, of the total releasable Ca 2ϩ , about 54% was released by ionomycin alone; the remaining 46% was then attributed to the Ca 2ϩ content selectively released from the acidic pool.
We functionally tested for the existence of InsP 3 receptors on granules. Fig. 1B shows that InsP 3 production, induced by the muscarinic agonist carbachol (CCh, 0.5 mM), reduced the Tgsensitive pool but had no effect on the size of the peak induced by monensin application. In fact, in the presence of CCh, the acidic pool represented 49 Ϯ 1.7% (n ϭ 7) of total mobilizable Ca 2ϩ . Finally, Fig. 1C shows that pretreatment with Tg abolished the peak in [Ca 2ϩ ] i induced by CCh, indicating that in this cell type, InsP 3 -and Tg-sensitive pools fully overlap. Altogether the data demonstrate that acidic Ca 2ϩ compartments in Ins-1 cells are depleted neither by InsP 3 produced by receptor stimulation nor by inhibition of Tg-sensitive pumps.
Capacitance Changes After Activation of Voltage Operated Ca 2ϩ Channels- Fig. 2 shows fast changes in Cm, membrane conductance (Gm) as well as series conductance (Gs) before and after a 200-ms depolarizing pulse from Ϫ80 mV holding potential to 0 mV. With 200-ms pulses, individual cells displayed ⌬Cm ranging from 20 to 400 fF with an average of 49.2 Ϯ 4.4 fF (mean ϮS.E., n ϭ 76). Increasing the pulse duration to 400 ms led to an increase in ⌬Cm of 60% when compared with a 200-ms depolarizing pulse in 4 of 9 cells (data not shown). However, these longer depolarizing pulses caused a rapid rundown of the evoked Ca 2ϩ currents. We therefore decided to perform the experiments with 200-ms pulse duration.
By following Cm at a lower frequency (3 Hz) it can be seen that upon such a depolarization ⌬Cm remained constant for up to 4 min before a rundown in secretion was observed, as long as these pulses were at least 20 s apart (Fig. 3A). This result indicates that secretion in Ins-1 cells is not easily "exhaustible" by successive 200-ms depolarizing pulses. With this protocol, maximum secretion increased the initial membrane capacitance by about 10% and was equivalent to fusion of approximately 300 granules (n ϭ 13) (assuming the mean diameter of the insulin-containing granules to be 250 nm and a membrane capacitance of 1 F/cm 2 ). However, since an interval of 20 s is too short to apply test substances between subsequent depolarizations, the interpulse duration was increased to 90 s. From the time course of ⌬Cm, we also noticed that, upon depolariza-tion, the majority of the cells (65%) displayed only an increase in Cm; in the remaining 35%, slow endocytotic processes were observed after the first and, occasionally, the second pulse (see Fig. 3B), whereas large, abrupt endocytotic events were never observed under our experimental conditions.
In addition to Cm and Ca 2ϩ current, the [Ca 2ϩ ] i was monitored as described under "Experimental Procedures." As can be seen in Fig. 4A, the amplitude of the [Ca 2ϩ ] i peaks decreased upon subsequent depolarizations, whereas the integrated [Ca 2ϩ ] i peaks and Ca 2ϩ currents (⌬Ip) (Fig. 4C) as well as ⌬Cm (Fig. 4B) remained unchanged. The decrease in [Ca 2ϩ ] i peak amplitude is probably due to the relatively slow influx of fura-2 from the pipette, leading to changes in the Ca 2ϩ buffering capacity of the intracellular medium during prolonged incubations (35).

FIG. 2. Cm changes induced by depolarization in Ins-1 cells.
Activation of a Ca 2ϩ current by a 200-ms depolarizing step from Ϫ80 to 0 mV (bottom panels) induces fast changes in Cm (upper panel), recorded as described under "Experimental Procedures." Monitoring Gs and Gm (second and third panels) shows that, after the depolarizing pulse, Gm transiently increases and returns to basal values within 100 ms, whereas Gs does not change. The figure also shows the time used to calculate the ⌬Cm (shaded area).

Role of Granular Ca 2ϩ Content in Secretion-
We first tested if drug application by itself induced ⌬Cm. From the low frequency recording of Cm it can be seen that during application of ionomycin, monensin, or the combination of the two, an increase in Cm occurred ( Fig. 5A-C, lower panels). Since similar increases were observed when the solvents ethanol or Me 2 SO were tested and when experiments were performed at room temperature to inhibit regulated secretion (36), we conclude that to a large extent the observed changes are due to the solvent. Chloroquine, on the other hand, being dissolved in Ringer's solution, had no effect. We also tested whether drug application changed the intracellular pH; therefore in some experiments BCECF was included in the intracellular solution instead of fura-2. These experiments showed that application of none of the drugs, applied alone or in combination, significantly altered the cytosolic pH when cells were kept in the whole-cell configuration (data not shown).
To determine the role of the granular Ca 2ϩ content in granule fusion, the increases in Cm in response to 200-ms depolarizing pulses were thus monitored immediately before and after application of the different drugs. Changes in Cm at the second and subsequent depolarizing pulses were normalized to the change obtained in the first pulse. When ⌬Cm was monitored in untreated, control cells after this normalization protocol, it remained stable during the second depolarization (Fig. 6, n ϭ  40). Figs. 5B and 6 show that, after a brief application of ionomycin, the subsequent depolarizing pulse caused a consistently larger increase in ⌬Cm (47 Ϯ 12%, n ϭ 10). The increase in ⌬Cm after the ionomycin pulse was relatively long lasting since it was maintained for at least 3 min and did not depend on larger Ca 2ϩ currents during subsequent depolarizations (compare Figs. 7, A and B).
Application of chloroquine between the first and the second pulse had no effect (n ϭ 6), whereas monensin led to a quite variable stimulation (Figs. 5A and 6). On average, however, the stimulation caused by monensin treatment between two successive depolarizing steps was not statistically significant (23 Ϯ 21%, n ϭ 9; see also Fig. 7).
A completely different pattern was observed when the Ca 2ϩ ionophore and chloroquine (or monensin) were applied together in order to discharge the Ca 2ϩ content of the acidic pools. The ⌬Cm increase following the depolarizing step, elicited after the discharge of the acidic Ca 2ϩ pools (Figs. 5C and 6), was reduced not only with respect to the potentiation caused by ionomycin (46 Ϯ 13%, n ϭ 17, monensin/ionomycin; 54 Ϯ 9%, chloroquine/ ionomycin, n ϭ 6) but was also reduced with respect to untreated control cells (21 Ϯ 9%, monensin/ionomycin; 31 Ϯ 2%, chloroquine/ionomycin).
The reduction in ⌬Cm following the depolarizing step was also observed with further test pulses, i.e. it was prolonged for up to 3 min (Fig. 7B). It is worth mentioning that there was no ] i and Cm changes were monitored as described under "Experimental Procedures." ⌬Cm is plotted as described in Fig. 3. C, the integrated Ca 2ϩ peak was obtained from the trace shown in panel A, whereas the Ca 2ϩ charge (q) was obtained by integrating the Ca 2ϩ current. significant change in either Gm or Gs when the second depolarizing pulse was compared with the first pulse either in controls or treated cells (data not shown). Moreover, application of ionomycin, monensin, or the combination of ionomycin and monensin had no significant effect on changes in [Ca 2ϩ ] i (⌬Ca) (Fig. 5) and ⌬Ip of the subsequent depolarizing pulses (Fig. 7A).
The inhibition of secretion did not exceed more than 30% of the control values. A possible explanation for this incomplete inhibition is that a complete alkalinization by monensin or chloroquine (and therefore complete discharge of granule Ca 2ϩ ) takes longer than the application time of 20 s used in our electrophysiological experiments. To test this possibility, studies were performed where the H ϩ exchanger was present for 5 min before the first depolarizing pulse was given, to ensure a complete breakdown of the pH gradient. After this prolonged incubation, secretion during the first depolarizing pulse was within the expected range of variability (43 Ϯ 7.2 fF; n ϭ 7). However, the prolonged treatment with monensin did not further increase the level of inhibition obtained when ionomycin was applied between the first and second pulses (24 Ϯ 12%, n ϭ 7; Fig. 6).
Insulin Secretion Studies-To determine whether the reduction in ⌬Cm caused by acidic Ca 2ϩ pool depletion was attributable, at least in part, to fusion of insulin-containing granules, we followed the release of insulin in populations of cells treated with protocols that mimic those used in Fig. 5. Cell suspensions obtained from monolayers were challenged with two pulses of 30 mM KCl of 1-min duration, 5 min apart. As summarized in Fig. 8, secretion during the second depolarizing pulse was 25 Ϯ 4% (n ϭ 3) of that obtained during the first challenge. This reduction in secretion probably reflects a reduction in readily releasable insulin granules, although a rundown of the Ca 2ϩ peak after depolarization may also contribute to this effect (34). One-min stimulation with 1 M ionomycin between the first and second KCl pulses resulted in a less drastic reduction of insulin secretion (62 Ϯ 24% that of initially released, n ϭ 3). When a combination of ionomycin and chloroquine (or monensin) was employed, secretion during the second KCl pulse was 25 Ϯ 11 and 24 Ϯ 4% (n ϭ 3), respectively; i.e. the potentiating effect of ionomycin was completely abolished. DISCUSSION In the ␤-cell line Ins-1, as in other secretory cells, a relatively high proportion of intracellular Ca 2ϩ appears to be stored in acidic structures. In fact, these cells respond with a large [Ca 2ϩ ] i increase to the protocol previously employed to reveal this compartment, i.e. the application of drugs that collapse internal acidic pH gradients (monensin or chloroquine) after addition of the Ca 2ϩ ionophore ionomycin. The subcellular localization of acidic Ca 2ϩ pools has not been determined with certainty, although it is likely that in Ins-1 cells, as in other cell types, it is heterogeneous. A rough estimation of the contribution of insulin granules to the Ca 2ϩ content of the acidic pool can be obtained by considering the total releasable Ca 2ϩ of Ins-1 cells (4.5 nmol/mg of protein, this work), the cell volume occupied by the granules (1.2%), 2 and the releasable granule Ca 2ϩ (about 125 nmol/mg of granule protein, Ref. 10). By using these parameters, intragranular Ca 2ϩ mobilization could be as high as 1.5 nmol/mg of protein. We have shown here that in Ins-1 cells 46% (i.e. 2.4 nmol/mg of protein) of the total releasable Ca 2ϩ is due to the acidic pool. Although based on a number of assumptions, these values indicate that insulin granules represent a major part (more than 60%) of the acidic compartment in this cell type.
The main goal of this investigation was to establish the role played by Ca 2ϩ trapped in the secretory compartment in the process of secretion. To address this question, we first investigated whether or not the acidic compartment could contribute to (i) the [Ca 2ϩ ] i changes induced in Ins-1 cells by the muscarinic agonist CCh and (ii) the [Ca 2ϩ ] i changes induced by depolarization. The finding that Ca 2ϩ mobilization induced by thapsigargin or InsP 3 production through activation of muscarinic receptors does not affect the acidic pool is meaningful. In fact, given that insulin granules represent a large proportion of that pool, it confirms by a functional approach the conclusion of Ravazzola et al. (18) that InsP 3 receptors are not expressed on the membrane of the secretory granules of ␤-cells. Similarly, a role for the acidic pool (and thus for insulin granules) in Ca 2ϩinduced Ca 2ϩ release is unlikely since the increase in [Ca 2ϩ ] i caused by depolarization was indistinguishable in controls and cells whose acidic pool had been depleted (Fig. 5).
We next tested the possibility that intragranular Ca 2ϩ plays a role in the secretory process by monitoring membrane capacitance in single Ins-1 cells under different experimental conditions. In untreated cells, the magnitude of ⌬Cm has a tendency to decrease during a series of successive pulses; however up to the fourth pulse (i.e. 300 s), ⌬Cm is fairly constant. On the contrary, manipulation of intracellular Ca 2ϩ in the time interval between the first and the second depolarizing pulses significantly changed the extent of secretion. In fact, depletion of Ca 2ϩ from nonacidic stores led to a prolonged stimulation of secretion up to 50%. Such a priming action of ionomycin has been described previously (4), but the fact that it can last for several min at resting [Ca 2ϩ ] i is a novel observation. Releasing Ca 2ϩ from the ionomycin-sensitive compartments may favor granule recruitment from a distant cytoskeletal-anchored pool (38) or by promoting priming of granules at a late, post-docking step (39). Such a priming has been previously described by mechanisms that cause long lasting phases of moderately elevated [Ca 2ϩ ] i (31).
In marked contrast with the potentiating effect of a brief increase in [Ca 2ϩ ] i , releasing Ca 2ϩ from the acidic compartments led to inhibition of secretion that reached 50% when compared with cells treated only with ionomycin. Since breakdown of the intracellular pH gradients by itself was without effect and the inhibition was observed with both monensin and chloroquine (two agents that act on pH gradients by different mechanisms), it can be concluded that the inhibitory effect is due to the release of Ca 2ϩ from the acidic organelles, including insulin granules. Since our alkalinization protocol is by no means specific for the granules, the question can be raised as to whether the reduction in secretion is due to the decrease in Ca 2ϩ within the granules themselves or in other acidic compartments (trans-Golgi network or lysosomes). The observation that inhibition is maximal within a few tens of seconds after FIG. 8. Effect of the Ca 2؉ -depleting protocols on insulin secretion induced by KCl in cell populations. Cell suspensions (10 6 /ml) were depolarized with two pulses of 1 min duration of 30 mM KCl applied 5 min apart in Ringer's solution. Insulin secretion during the second KCl challenge was measured by radioimmunoassay as described (52) and was normalized to that obtained during the first pulse. 1 M ionomycin alone or a combination of 1 M ionomycin with 2 M monensin or 40 M chloroquine in Ca 2ϩ -free Ringer's solution containing EGTA (1 mM) was applied for 1 min between the first and second depolarizing pulses. * p Ͻ 0.05; 2-tailed Student's t test.